Three complete audits, start to finish. Read one before you spend a dollar.

AI Operations Audit: a branding and marketing agency

Iterategy · Prepared for Windrow Collective (SAMPLE, a fictional business) · 2026-09-01

Turnaround: 7 business days · Order SAMPLE-WINDROW

1. Read this first

This sample audit describes a fictional six-person branding and marketing agency in Boise, invented so that a real buyer can see exactly what she would receive. Every person, client and number in it is made up. The method, the sources and the arithmetic are real, and they are the same ones a paying order gets.

What this is. You bought a look at where your week goes and what it costs, with a number on every finding and the arithmetic printed beside it. This is that document. You can disagree with any number in it; the working is shown so that disagreeing rescales the answer rather than throwing it out.

What we found, in the order we would fix it.

1. You build two client reports by hand every month, and you already pay $99 a month for the tool that builds them. The dashboards for those two clients were never set up. Worth $7,837 a year.
2. You chase overdue invoices yourself, and your accounting plan already has reminders built in and switched off. Worth $6,899 a year in your time.
3. Retainer hours get retyped from one place to the next and the tracker is wrong within days. Worth $4,987 a year.
4. About 16 hours a month of work you deliver is never invoiced, because the record it would be billed from does not exist. Worth $4,176 a year, and it is the same fix as number 3.
5. Every proposal gets rebuilt from an old document while a $49 a month proposal tool sits unused. Worth $3,571 a year.
6. That $99 and that $49 are $148 a month you already spend on two tools that do two of the jobs above. Counted at nothing, because it pays for the plan rather than being a saving on top of it.

What matters most, and why it is first. The reporting one. It is not the biggest number by much, but it is the only one where you are already paying a monthly bill for the exact tool that does the job and nobody ever finished setting it up. We checked every channel those two clients run, one at a time, against the tool’s live connector list on 2026-09-01. All seven are supported. There is nothing to buy, nothing to migrate and nothing to decide. It is an afternoon.

What it adds up to. 6.8 hours a week of yours and 2.6 of your team’s, and $27,470 a year. Even counting only what we’re certain of: $19,723/yr.

And the whole plan is paid for by money you are already spending on tools you do not use. Not one recommendation in this report asks you to spend a dollar you are not spending today. The one place that could have needed money, a trustworthy record of who worked which hour, turned out to be a setting inside the accounting plan you already have.

One thing we will not round up. You asked what you would do with ten hours a week back. We found between five and eight of yours, and our arithmetic lands at 6.8. Half of that is the reporting and the invoice chasing. It is not ten. We would rather hand you a number you can hold us to than the one you asked for.

And one answer you did not expect. You told us two clients pay 30 to 45 days out and that it is just how those companies pay. You are right about one of them and wrong about the other, and their twelve invoice dates show which is which. It is in finding 2.

What happens next. Section 6 is a week-by-week plan you can start on Monday, free things first. Your questions are answered free for two weeks: reply to the email this arrived with.

2. What we looked at, and how

What you gave us. Your intake form, submitted 2026-09-01, all 34 answers. A follow-up question set answered in your own words the same day, covering retainer terms, a typical week’s hours, your aging report, twelve dated invoice rows for the two clients who pay late, your reporting and proposal setup, your plan names, a count of out-of-scope asks, and what your people cost. Three things you sent over: your proposal template, one client report as delivered, and your Master Tracker export as it stood, not tidied first. You offered read-only access to your accounting and your task tool; this audit did not need to use it, and the offer is still open if you want any number below re-derived from the source.

What we did not look at, because you asked us not to. Your CRM. You said the prospects in it do not know they are being discussed. Nothing in this report depends on it.

What we counted, and how you can re-count it. Every count below can be re-run by you.

How the money was worked out. Hours times a loaded hourly rate times 50 weeks. The rates are in Section 7, all three of them, with their sources. Where your own payroll figure was lower than the published Boise figure, yours was used. That happened once and it took about $1,300 a year out of the largest finding.

As is, from inquiry to cash

Someone finds you by referral or on a professional network and books an intro call through your booking link. If it is a fit, you rebuild a proposal from an old document, usually late at night. They say yes by email. You open a project in your task tool and bring in contractors. Work happens, you approve it, you deliver. At month end or on project completion you create the invoice in your accounting system and email it. Then you wait, and follow up once, sometimes three or four times.

Four places the same fact is typed twice or more.

1. Hours worked, from your task tool’s time entries into the Master Tracker.
2. The Master Tracker figure into the invoice.
3. Platform numbers, screenshotted out of six ad and analytics tools into two slide decks.
4. Project status, out of the task tool and into client-facing recaps.

Three places work stops.

1. A client asks for something extra. It gets delivered. It does not reach an invoice unless you personally remember. About 16 hours a month go this way.
2. An invoice goes out. Nobody chases it but you, and only when you notice.
3. Retainer invoices for one client go out around the 10th of the following month, for work finished up to six weeks earlier.

The eleven manual jobs running weekly or more

Rows 1 to 4 total 14.5 hours a week, 10.5 of them yours. That is the universe this report draws from. Row 5 is client work rather than admin, shown separately so it is not counted twice. Row 8 is your bookkeeper’s whole engagement and is counted at nothing throughout.

3. The six findings, in payback order

Ranked by return per hour of your effort, not by size. Every finding carries the arithmetic, a one-word verdict, and the dated source that settles it.

Finding 1. Two clients get hand-built reports while the reporting tool you pay for sits with their dashboards never built

In your words: “AgencyAnalytics for reporting, which I pay for but haven’t fully set up”, and “two client dashboards never got built”.

What it costs today.

• You: 2.5 hrs/wk x $57.49 x 50 = $7,186
• Your coordinator: 2.0 hrs/wk x $26.10 x 50 = $2,610
• Total: $9,796 a year to hand-build two monthly reports.

Recoverable at 80%: $7,837 a year. That returns 2.0 of your hours a week and 1.6 of your coordinator’s.

Verdict: configuration. Not a product limit, not a habit. The tool does this job and it was never switched on for these two clients.

What settles it, checked one channel at a time on 2026-09-01. The question that decides this finding is whether the tool can actually connect to what those two clients run. Both clients’ channels were checked against the vendor’s live integration pages:

All seven are native connectors. Had one failed, this finding would have changed shape and we would have said so.

What you already own and are not using. AgencyAnalytics, entry plan, $99 a month. Unused feature: the two client dashboards themselves, plus scheduled delivery. Four dashboards are already built and working for four other clients, which means you have done this exact task four times and it works.

The fix, and how long it takes. About three hours of your coordinator’s time, once.

1. Connect the four sources for Client A and the three for Client C. Note that the Meta Ads connection for Client A failed twice in early 2025 and you gave up. That platform’s permissions have changed since; treat it as a fresh connection, not a retry.
2. Copy the layout from one of your four working dashboards.
3. Add a text box for your commentary and an image widget for the creative that ran. Images take SVG, PNG or JPEG up to 1MB.
4. Put your logo and colors on it, and set it to send on a schedule.

What the tool will not do, and this is the 20% we did not claim. It will not write your three commentary slides; it hosts them, you still write them. And it will not reconcile spend against what each client was invoiced. That is your two reconciliation slides, and the only route is typing the invoiced figure into a text box each cycle. We did not check whether the vendor has an accounting connector that would close that gap, and we are not going to assume one.

Before you build: read your billing page. Your $99 does not match any plan this vendor publishes today or has published in the past. They replaced their three tiers with a single per-client plan between 2026-04-18 and 2026-05-15, at $20 per client per month billed annually or $25 billed monthly. You are most likely on a retired plan. Two things follow. First, check your client allowance covers six, since you have four dashboards and want two more; your own note says it allows more than you are using. Second, do not migrate. Six clients on the current plan is $120 or $150 a month, against the $99 you pay now. Staying where you are is cheaper.

Finding 2. Overdue invoices are chased by hand, by you, and the reminders you already pay for are switched off

In your words: “I’m the one who has to notice it’s overdue and chase it, there’s no automatic reminder happening.” And, separately: “What I actually want is something that chases the invoice for me so I never have to think about it.”

What it costs today.

• You: 3.0 hrs/wk x $57.49 x 50 = $8,624 a year
• Your bookkeeper’s half hour a week is not counted. Freeing it does not change what you pay her.

Recoverable at 80%: $6,899 a year. That returns 2.4 of your hours a week.

Verdict: configuration, with one honest limit. Automatic payment reminders are on your accounting plan and are off today. You can turn them on this afternoon. But there is something the software cannot do, and you told us it matters, so we are telling you before you find out.

The limit, stated plainly. You also wrote: “I don’t want anything that auto-sends a client email without me seeing it first, at least not yet.” Your accounting system cannot hold a reminder for your approval before it goes out. That mode does not exist, on your plan or on any plan. Its help documentation has exactly two settings, automatic and manual, and manual means finding each invoice yourself. The desktop version of the same product does have a “Review and send” queue; the online version has no equivalent. Upgrading to the top tier at $340 a month does not buy it either. We checked all three on 2026-09-01.

So the fix is the other shape, and it still answers both of your sentences.

Approve once, then run, with a human escalation lane. Four parts:

1. You write the words. All three reminder texts, subject line and body, in your own voice. Both are fully editable and you can drop in the invoice number and the client’s name automatically. Nothing goes to a client in language you have not written and approved. That is what “seeing it first” becomes: you see every word before any of them exists, rather than one at a time. You turned a reminder on for one client a year ago and killed it the same day because it read like a collections notice. It read that way because it was theirs. This one is yours.
2. You watch it for two weeks first. Days 1 to 14, leave the automatic setting off and send those exact three texts by hand from the invoice list. Same words, same timing, your finger on the button. You see them land on real clients. On day 15 you switch the toggle. Your own sentence was “at least not yet”, so this is the “yet”.
3. Schedule: three reminders, no more. Reminder 1 three days before due. Reminder 2 three days after. Reminder 3 ten days after. Your plan supports up to three, at any offset up to 90 days.
4. The fourth touch is you, on the phone. Anything still unpaid after reminder 3 is a conversation, not a message. That is client relationship communication and this report does not automate it.

Two things the settings will not do, so you can decide with your eyes open. The templates are global, so you cannot give one client a gentler cadence or exempt them. And changes are not retroactive; they apply to future reminders only.

Now the question you asked us not to ask. You told us two clients run 30 to 45 days out and that it is just how those companies pay, not something on your end. You are right about one of them and wrong about the other. Here are your own twelve rows.

Client B: you are right. There is nothing to fix and we are not going to invent something.

You invoice them on the first of the month, every month, without fail. They pay in 44 to 46 days, mean 44.8. Their accounts payable department told you they run a 45-day cycle regardless of what the invoice says, and six rows agree with them. Reminders will not move this and you should expect them not to. When reminder 3 lands on this client every month and nothing happens, that is not the setting failing.

Client A: about ten and a half days of it is yours.

They pay in 40.8 days on average, which is genuinely slow against your net 30. But look at the left column: you invoice them on day 10.5 of the month, on average, for the previous month’s work. Every month, not only the slammed ones. That ten and a half days sits before the invoice exists, so no client behavior causes it and no reminder can shorten it.

Counted end to end, this client is your slowest, not your second slowest. Work done mid-month waits about 15 days for the month to end, another 10.5 for the invoice, then 40.8 to be paid. Client B, who looks worse, is faster from work to cash because you invoice them on the first.

The fix for that half is a calendar entry, not software. Invoice the retainers on the last working day of the month they cover.

The cash this releases, shown separately because it is not the same kind of money. Ten and a half days of Client A’s $84,000 a year is $2,416, once. If your other six retainers run on the same rhythm, and your own note that “when I’m slammed the retainer invoices go out a week or two into the next month” suggests they do, the total is up to $15,189, once. This is not in the $27,470. It is cash arriving earlier, not cash arriving that was not coming. It happens one time, in the month you change the rhythm. The recurring part of this finding is your 2.4 hours a week and that is what the number above counts.

Finding 3. Retainer hours are retyped from one system to the next, and the tracker is wrong within days

In your words: “Retainer hours get tracked in Asana time entries, then I re-enter a summary into the Master Tracker spreadsheet, then it goes into the invoice.” And: “I update it by hand and it’s usually a few days behind reality.”

What it costs today.

• You: 1.5 hrs/wk x $57.49 x 50 = $4,312
• Your account lead: 1.0 hrs/wk x $38.45 x 50 = $1,923
• Total: $6,234 a year maintaining a record you told us nobody trusts.

Recoverable at 80%: $4,987 a year. That returns 1.2 of your hours a week and 0.8 of your account lead’s.

Read findings 3 and 4 together. They are one fix and two different pools of money. The tracker drifting and the work that never gets billed have the same cause: there is no trustworthy record of who worked which hour for which client. Finding 3 counts the labor of maintaining the bad record. Finding 4 counts the revenue the bad record loses. They do not double count, and they are not two projects. The 30-day plan treats them as one.

Here is the bad record, in your own export. Hours sold, hours per the tracker, hours per the task tool:

Three of those rows are worth a closer look. On client C the tracker says you delivered exactly the 35 hours you sold, to the hour, and the time entries account for 12. On client G, exactly 50 sold and exactly 50 used, with 26 logged. Two clients landing exactly on their number is not measurement, it is the month-end reconstruction you described. The row for client H was last updated on 30 July.

Verdict: product limit, and not the one you would guess. Your task tool’s Starter plan does not have native time tracking. Its pricing page carries a line reading “Time tracking with integrations” in the Starter column, which means you can connect a separate time tracking product to it. Actual logging of time against a task is on the Advanced plan and above; the vendor’s own help page lists the plans that have it and Starter is not among them. Verified 2026-09-01.

So the “half filled in on a good month” time entries are not a discipline problem. They are people typing into a field that was never built for it.

The fix, and it costs nothing. The single cheapest path is a setting inside software you already pay for.

Your accounting plan already does this. Built-in time entry, with a Customer field, a Billable checkbox and a rate per hour, is included on your tier and on the tier below it, at no extra charge. It produces reports for time by customer and for unbilled time, and billable time flows straight onto the invoice. Intuit’s own help articles, updated 2026-08-03 and 2026-08-05, read 2026-09-01.

1. Turn on time tracking in the accounting settings.
2. Add each retainer client with its monthly hours.
3. Three people log hours there weekly. Same keystroke your account lead already spends on the tracker, one step later in the chain instead of one step earlier.
4. Retire the Master Tracker. The hours column stops existing. Renewal dates can stay in a sheet.

The one thing to check before you start, because it is the only thing that could stop this: the vendor documents that the weekly timesheet does not work properly if you run their own payroll product alongside it. If you do, tell us and we will re-cost this in your two-week question window at no charge.

The three paths, priced, so you can see why the free one won.

The free path wins on cost and it also wins on the thing you actually want, because it puts the hours in the same system that produces the invoice. The other two still leave a gap between where time is logged and where it is billed, which is the gap this finding is about.

Finding 4. About 16 hours a month of delivered work is never invoiced

In your words: “Between ‘client asked for something extra’ and ‘that extra thing gets billed,’ it almost never gets billed unless I personally remember to add it.” And: “That wasn’t a decision, it’s just what happened.”

How much, worked out two ways that agree.

From your Master Tracker export. Add up only the clients who used more hours than they bought: Client A 7 over, Client D 1 over, Client F 8 over. 16 hours in that month. Under-use on other clients is not subtracted, because hours you did not spend on one client are not billable to another.

From your own count. About 2 out-of-scope asks per client per month across 8 retainers is about 16 a month, mostly an extra revision round or a quick graphic, of which about one a quarter reaches an invoice. At roughly an hour each, that is 16 hours a month.

Two routes, one from a spreadsheet and one from memory, land on the same number. 16 hours a month, 192 hours a year.

What it is worth.

• 192 hours a year delivered beyond the retainer
• Six of your eight agreements carry an overage clause, so the eligible pool is 192 x 0.75 = 144 hours
• Captured at 20% = 28.8 hours
• At your blended retainer rate of $145 an hour = $4,176 a year

Recoverable: $4,176 a year. No hours come back to you here. This finding bills for hours already worked.

Why 20% and not more. The cap for this kind of finding is 30%, which would be $6,264. We did not claim it. You personally approve every client communication and do not want anything sent without you. So every hour this makes visible still has to pass through the same decision the finding is about. A system that surfaces the hour does not guarantee you bill it. Being right about the rate and cautious about the volume is a better trade than being optimistic about both.

And why the rate is not discounted. Normally we would count recaptured revenue at your margin, because winning revenue back usually means doing the work again. Not here. The work is already done. The designer already ran the extra revision round, the contractor is already paid, the client already has the file. Billing for it adds cash and adds no cost, so it counts at your full rate.

Verdict: habit, with a contract problem sitting underneath it for two clients.

The part that is not a software finding at all. Six of your eight retainers are on the 2022 agreement with its “$150 per hour with prior client approval” clause. The two oldest are on a letter that says the retainer covers “reasonable additional requests”, with no rate. For those two clients, no amount of time tracking recovers a dollar. There is nothing to bill against. The fix for them is a contract conversation at renewal, not a setting.

That is why this finding counts six eighths of the pool and not all of it.

One number you can improve on, and we cannot. You told us one of those two legacy clients is also one of the chronic late payers, which does not tell us which two of the eight they are. Of the 16 overage hours in your export, 7 sit on Client A, 8 on Client F and 1 on Client D. If the two legacy agreements turn out to be A and F, this finding is worth a few hundred dollars, not $4,176, and the real answer is a contract conversation with two clients. You can settle it in a minute and we would rather you knew that than found out later.

The fix, for the six clients where there is something to bill. It is finding 3’s fix, and it needs nothing added.

1. Hours go into the accounting system with the Billable box ticked and the client selected.
2. On the 25th, run the unbilled time report. Five minutes.
3. Anything over the retainer, you decide: bill it, or waive it deliberately. Waiving is fine. It is waiving without knowing that costs money.

What this does not do, on purpose. It does not send anything to a client, it does not add a line to an invoice by itself, and it does not decide what to charge for. It puts a list in front of you once a month.

Finding 5. Every proposal is rebuilt from an old document while the proposal tool you pay for goes unused

In your words: “a proposal software subscription ($49/month) that I keep paying for but don’t actually use because I still write proposals from an old Google Doc”, “usually staying up late to do it”.

What it costs today.

• You: 3.5 hrs/wk x 50 = 175 hours a year on proposals
• Two thirds of that is assembling, not deciding. Your template is 9 pages: cover, about the agency, the client’s situation, scope, timeline, investment, team bios, three case studies from a bank of six, terms. Your own words: “roughly two thirds of the pages are the same every time.” And your own time split says about an hour and a half of a 4 to 6 hour proposal is the thinking.
• Assembling: 116.7 hours a year of yours, 16.7 of your team’s

Recoverable at 50%: $3,571 a year. That returns 1.17 of your hours a week.

Why 50% and not 80%. Most of a proposal is scoping and pricing judgment. That is the work and it does not template away. Claiming 80% here would be claiming we can automate your thinking, which is wrong and is the exact thing you told us not to touch. We are counting the third rewrite of the same capabilities paragraph, not the decision about what to sell.

One note on your own numbers. Your weekly figure of 3.5 hours implies about 3.2 hours per proposal, which is below the 4 to 6 you gave per proposal. We used the weekly figure because it is lower. Your per-proposal figures would have put this finding $1,000 to $3,000 a year higher.

Verdict: habit, sitting on a migration nobody did.

What you already own and are not using. Proposify, Team plan, $49 a month, one seat. Verified against the vendor’s live pricing page on 2026-09-01, and your $49 matches their current published rate exactly. All three features this fix needs are on it:

• Templates. “All accounts allow unlimited templates.”
• A content library. “Create and reuse sections, text snippets, and images that can be pulled into documents. Useful for pulling in case studies, team bios, and other bits of modular content.” That sentence is your six case studies and your team bios, described by the vendor.
• Electronic signature. “All plans allow you to get documents legally e-signed.” Today your clients reply “approved” by email. A signature is better evidence and it is included.

The fix. One evening, once.

1. Load the 9-page document as a template.
2. Break the never-changing pages into library blocks: about the agency, team bios, terms, and each of the six case studies as its own block.
3. Leave the four pages that get rewritten every time as blank sections. Those are yours to write.
4. Turn on the signature block instead of waiting for an email.

Next proposal: pick three case studies, write the four pages that are actually about this client, send. The two thirds you retype stops being typed.

One correction to what you may believe about your billing. The Team plan is billed quarterly or annually, not monthly. There is no month-to-month option on it. Your $49 a month is really $147 a quarter, committed a quarter at a time. That does not change what you spend; it changes when you can change it.

Finding 6. $148 a month buys two tools that do two of the jobs above, and neither is switched on

In your words: the paid-and-barely-used answer, and then your condition: “anything new has to replace the AgencyAnalytics or Proposify money, not sit on top of it. If the fix is to actually use those two, that’s the answer I was hoping for.”

It is the answer.

Counted at $0, and here is why that is not a technicality. $99 plus $49 is $148 a month, $1,776 a year, and it would be easy to put that in the total as found money. It is not found money. Finding 1’s fix is to use the first tool. Finding 5’s fix is to use the second. If you use them, the $148 is not saved. If you cancel them, findings 1 and 5 need a different fix or stay broken. The two paths are mutually exclusive and only one of them can be in a total. Counting both would be counting the same $1,776 twice.

So it is not a saving. It is the funding line, and it is the reason this whole report costs nothing to act on. Your condition was that nothing net-new sits on top of the $148. Nothing in this report is net-new at all. The one place that could have needed money, finding 3, turned out to be a setting in software you already have.

Verdict: configuration, twice.

One thing we found that you can have for free, if you want it. The three features finding 5 needs, templates, the content library and the signature, are all on the proposal tool’s cheaper plan, at $19 per seat per month billed annually or $29 billed monthly, against the $49 you pay. That plan allows 10 document sends a month and you send 4 to 6. If those three features are your reason for being on the more expensive plan, you are one tier too high, and moving down at renewal is $20 to $30 a month, so $240 to $360 a year.

We are not counting it, for the same reason as above, and we are not pushing it. Use the tool first for a quarter. If you end up wanting the pricing catalog or the integrations that the higher tier adds, stay. If you do not, move down at renewal and keep the difference.

4. What we deliberately did not pursue

Everything we looked at and left, so you can see it was decided rather than missed.

• Anything that sends a client email without you, and anything touching creative direction. Your standing rule, and it is a good one. It bound findings 2, 4 and 5, and it is why finding 2 is designed the way it is rather than the easy way.
• Your CRM, and prospect automation. You declined access because the prospects in it do not know they are being discussed. We do not model what we have not been allowed to see, and no finding here depends on it.
• Replacing your accounting system. Nothing in your intake asks for it, and the migration would cost more than every finding in this report combined.
• Replacing your task tool. Same, and the retyping problem follows you to any tool if the time record stays optional. That is why finding 3 moves the record rather than the tool.
• Your design software. Used well, by your own account, and no finding would clear our bar.
• Your assistant for content drafts. Already working and already saving you real time. An audit that recommends what you already do is padding.
• Re-entering project status into client recaps (process 9). Real, and it is a communication task rather than a data task. Automating the recap means writing to a client on your behalf, which is the line you drew.
• Your bookkeeper’s five hours a week (process 8). Bookkeeping is the work, not overhead, and freeing half an hour of her invoice chasing does not change what you pay her. Her half hour is counted at nothing in finding 2, which lowers that finding by about $600.
• Renegotiating or firing the two chronic late payers. We answered whether the lateness is theirs or yours, which is what you paid for. Which clients to keep is your call and we are not going to make it from a spreadsheet.
• Your contractor bench and how work is assigned to it. No signal in your intake and nothing to size it with. If you want it looked at, it is a separate conversation.
• Migrating your reporting tool to its new plan. We checked, and it would cost you more. Named here so you know it was considered.
• Anything net-new that does not come out of the $148. Your condition, and in the end nothing in this report needed money at all.

5. What this is worth, and the bar we hold ourselves to

Total identified opportunity: 9.3 hours a week and $27,470 a year.

New tool spend: $0 a year. Net: $27,470 a year.

Net equals identified because nothing here spends money. That is not a rounding; it is the shape of this business’s problem. The three largest findings are all switching on capability you already pay for.

Even counting only what we’re certain of: $19,723/yr. That is findings 1, 2 and 3, the high-confidence ones. It excludes finding 4, where the volume is solid but the capture is a judgment, and it excludes finding 5, where the finding is certain and the hours are your estimate.

The hours, in your units. 6.8 of yours a week and 2.6 of your team’s. Your hours figures are estimates, by your own label, so at plus or minus 20% the honest range on yours is 5.4 to 8.1. You asked for ten. We found five to eight, and we are not rounding it up.

The cash release, which is not in any number above. Moving the Client A retainer invoice to the last working day of the month it covers releases $2,416 once, and up to $15,189 once if your other six retainers run the same rhythm. That is cash arriving earlier, not new cash, and it happens one time. It is reported separately because mixing a one-time balance-sheet gain into an annual figure would inflate this report by more than its largest genuine finding.

The bar we hold ourselves to. We do not send an audit that finds less than $4,500 a year against a $1,500 fee. This one finds $27,470, and $19,723 counting only the high-confidence findings. That is a standard for our own work, checked before this document reached you.

6. The first 30 days, in order

Free and fast first. Nothing in this plan costs money.

Every step below costs $0, so there is no cost column: there would be nothing in it. The time column is your time, which is the only thing any of this spends.

Total cost: $0. Total return: $27,470 a year, plus $2,416 to $15,189 of cash, once.

If you only do three things.

1. Build the two dashboards. Week 2, three hours of your coordinator’s time, $7,837 a year. The highest return per hour of effort in this report by a distance.
2. Move the retainer invoice run to the last working day of the month. Week 1, fifteen minutes, and it is the only line here that puts cash in the bank rather than time on your calendar.
3. Turn on time tracking in your accounting system. Week 3, one hour, and it is the single fix behind two findings worth $9,163 a year between them.

One sequencing note. Week 3’s reminders depend on week 1’s texts existing and on two weeks of you sending them by hand. Do not skip week 1 and switch it on directly. The two weeks are not caution for its own sake; they are how you find out that the wording you wrote reads the way you meant before a client reads it.

And one thing not to do. Do not run the Master Tracker and the new time record side by side “for a while to be safe”. Two records that disagree is exactly the problem you are paying to fix, and the month you spend keeping both is the month nobody trusts either.

7. Appendix: our assumptions

Every judgment sitting on top of the numbers, so that disagreeing with one rescales the rest instead of discrediting the report.

The rates

Sources. Wages: US Bureau of Labor Statistics, Occupational Employment and Wage Statistics, May 2025 estimates for the Boise City, ID metropolitan area, released 2026-05-15 as USDL-26-0725. Benefits load: US Bureau of Labor Statistics, Employer Costs for Employee Compensation, March 2026, released 2026-06-12 as USDL-26-0827: private industry total compensation $46.60 per hour worked against wages and salaries of $32.60, so 46.60 divided by 32.60 is 1.429. Both read 2026-09-01.

Where your figures and the published ones disagreed, the lower one was used. Your own numbers would have put you at $62.40 an hour and your senior designer at $38.58. The published Boise figures were lower in both cases, so those were used. For your coordinator the published figure was $42.86 and yours was $26.10, so yours was used. That one choice took about $1,300 a year out of finding 1.

Two rates we rejected, so the choice is visible. Art Directors in Boise gives $58.45, within a dollar of the rate used, which is a useful check. Marketing Managers gives $109.91 and was not used: Boise’s 780 marketing managers sit mostly in large corporate employers, and applying that to the owner of a six-person shop would nearly double the largest input in this report.

One vintage note. The wage data is May 2025 and the benefits factor is March 2026. There is no more recent Boise wage release and no Boise-specific benefits factor. Wages have risen since May 2025, so these rates are slightly conservative.

Working week and volumes

• 50 working weeks a year, not 52. Two weeks off.
• 2,080 hours a year for converting a salary to an hourly rate.
• Every hours figure in this report is your estimate, given in your own follow-up answers and labeled by you as estimates. Your intake form carried no hours at all. This is the largest single uncertainty in the report: five of six findings have an hours multiplier and all of them are yours. Correct any of them and we recompute free inside your two-week window.
• Volumes taken from your own documents rather than estimated: the 16 hours a month of overage, from your Master Tracker export. The twelve invoice dates. The aging summary. Those are the strongest numbers here.
• Two intake fields were not used at all. You answered 18 to “jobs or orders per week” and $6,500 to “average job or order value”. Multiplied out that is about $6M a year against your stated $500K to $1M. You told us 18 is deliverables shipped and $6,500 is a monthly retainer, which resolves it. Neither number appears as a multiplier in any finding. Our form asked an agency a question written for a trades business, and that is our problem to fix, not yours.

Method

Findings came from your intake and your follow-up answers, sized in a worksheet where every number shows its arithmetic, then written up. Every capability claim about a product was checked against that vendor’s live documentation on 2026-09-01 and is dated in the finding. Nothing here rests on what a tool used to do.

Three claims are worth naming as the ones we checked hardest, because a report is only as good as the thing it would have been wrong about:

1. Every one of the seven advertising and analytics channels your two clients run is a supported connector. Checked one at a time. Had one failed, finding 1 would have changed shape.
2. Your accounting system cannot hold an invoice reminder for your approval before sending, on any plan. The desktop product has that feature; the online product does not. This is why finding 2 is designed as approve-once rather than review-then-send.
3. Your task tool’s Starter plan has no native time tracking, despite a line on its own pricing page that reads as though it does. Confirmed on the vendor’s help documentation.

What we could not verify, listed rather than buried. Your $99 a month does not match any plan your reporting vendor publishes now or published before; your figure was used and the action is to read your billing page. We did not check whether that vendor has an accounting connector that could automate your spend reconciliation. We do not know whether you run your accounting vendor’s payroll product, which is the one thing that could block finding 3’s fix. And we do not know which two of your eight clients are on the legacy letter.

How this audit was produced

Data handling

This is a sample audit about a fictional business, so there is nothing of anyone’s to delete. On a real order: we would have been given read-only access to your accounting system and your task tool, that access is revoked on delivery day, and everything you shared is deleted within 30 days with a written confirmation. Your submitted answers never enter our source control at any point.

How to get help

Your questions are answered free for two weeks: reply to the email this report arrived with. We will check in on day 10. And a live 30-minute call to walk the findings together is included with your audit; the booking link is in that same email.

Bonus A: Getting named when a customer asks an AI assistant

A dated, honest read on whether AI assistants name your agency when someone asks for one.

What this section is, and what it is not. When someone asks an assistant “who are the good branding agencies in Boise”, the answer is assembled from what those systems can find about you on the open web. This section is a checklist of what makes you findable, and the fixes for each gap. It is descriptive, it is dated, and no dollar figure is attached to it. It is not in the $27,470 and it is not in the bar in Section 5, because we have no honest evidence for what an AI-sourced inquiry is worth to a marketing agency and we are not going to invent one.

Nobody can guarantee placement in an AI answer, including us. Anyone who tells you otherwise is selling something. These systems are probabilistic: the same question asked twice can give different answers, and they change without notice. What can be done is unglamorous and durable: make the facts about you easy to find, consistent everywhere, and machine readable. Those are fundamentals, not a trick, and they help ordinary search too.

For this sample, the honest result is nothing. The subject of this report is invented. No assistant can name a business that does not exist, so there is no snapshot to show you and we are not going to fabricate one. On a real order this section carries the checklist below filled in against your actual web presence.

What we check

What to do about a gap

Every gap gets a fix with three ways to do it: yourself, hire anyone, or hire us at a fixed price. None of these are dollarized, for the reason above.

For an agency specifically, two of these are worth more than the rest. Your structured data, because your service list and your city are the two facts an assistant needs to include you in a local answer. And answer-first structure on your service pages, because you sell to people who ask questions in sentences.

Method notes. The list of assistants worth checking, the crawler names, and which model sits behind which product all change on a scale of months. Re-run this quarterly. It is a snapshot with a date on it, not a permanent score.

Bonus B: A starter AI policy

A courtesy section, because the most common thing that stops a small team using these tools is not knowing what is safe to put in them. This is a starting point, written around your actual tools. It is not legal advice. Have a lawyer look at it before it becomes policy, especially the client-data section, since your client contracts may have their own confidentiality terms.

1. Approved tools

Adding a tool. One person approves it, which is you. The question to ask is where the data goes and whether it is used to improve someone else’s model. If the answer is not on the vendor’s page in plain words, that is an answer.

2. What can and cannot go into an AI tool

Safe: your own published work, public information about a client’s market, your own internal notes, anything already on a public website, invented example data.

Never: client login details of any kind. Unpublished campaign plans or launch dates. A client’s revenue, spend or margin figures. Contract terms and rates, including your own. Personal data about a client’s customers. Anything a client has marked confidential. Your own payroll figures.

The test, when it is not obvious: would you be comfortable if this text appeared in a competitor’s answer to a similar question next month? If not, it does not go in.

3. Client data

Ask first, and ask in writing. Some clients will be delighted and some have contracts that forbid it. A one-line email at the start of an engagement settles it for the whole relationship.

Default when you have not asked: no client data goes into a general purpose AI tool. Strip names, figures and identifying details before using a tool to help with the writing.

4. Acceptable uses

Drafting and rewriting your own words. Summarizing long documents you already have the right to read. Generating options to react to. Checking your own copy for clarity. Explaining something you do not understand yet. Producing first-pass structure for a proposal, a brief or a report, on the understanding that the thinking is still yours.

5. Prohibited uses

Sending anything to a client that a person has not read end to end. Presenting generated work as original research. Using it to make a decision about a person, including hiring. Putting anything from the “never” list above into any tool. Using generated imagery of real people without saying so. Relying on a factual claim from an assistant without checking it against a real source, which is the same rule this report holds itself to.

6. When something goes wrong

If confidential information goes into a tool by accident: tell one person immediately, which is you. No blame. The first hour matters more than the explanation. Then, in order: check whether the tool has a delete-conversation and an opt-out-of-training setting and use both. Write down what went in and when. Decide whether the client needs to be told, and lean towards telling them. Note what made it easy to do by accident and change that.

A near miss is worth reporting too, and it is the cheapest information you will get about where this policy is unclear.

7. Review

Every six months, and whenever you add a tool. Fifteen minutes: are the approved tools still the tools we use, has any vendor changed its data terms, did anything go wrong, does the never list need another line. Put the date of the last review at the top of the document so it is obvious when it has been forgotten.

AI Operations Audit: Cobble Creek Plumbing and Drain

Prepared by Iterategy · 2026-09-01 · Turnaround: 7 business days · SAMPLE AUDIT, order SAMPLE-COBBLECREEK

1. Read this first

This sample audit is built on a plumbing company we invented, Cobble Creek Plumbing and Drain, so we can show you what the real thing looks like without spending a customer’s numbers to do it. Everything about the business is made up: the owner, the staff, the call log, the quote sheet. Everything about the outside world is real, dated, and sourced so you can check it: the wage data, the profit margins, the software prices, what each tool can and cannot do.

What a customer buys is this. We read your intake form, we ask for the numbers we are missing, we work out where your time and money are going, and we hand you this document. Every finding carries a dollar figure and the arithmetic that produced it, so you can disagree with any single number and rescale the rest yourself instead of throwing out the whole report.

What we found, in the order we would do them.

1. Big quotes go quiet and nobody owns chasing them. About 60 large quotes a year, worth roughly $359,700, get no answer either way. Worth $11,613 to $23,225 a year.
2. You check Housecall Pro against QuickBooks by eye every week or two. Around 3,000 invoices a year, compared by hand. Worth $4,220 to $13,100 a year.
3. After-hours callers reach voicemail, and some of them hire somebody else by morning. Your own call log says about two and a half a week, confirmed. Worth $8,438 to $16,031 a year.
4. Payroll gets redone by hand on most runs. Worth $203 to $813 a year, and the reason it happens is not the one you have been assuming.
5. Comfort Club members lapse with nothing chasing the renewal. 24 lapsed over the last year. Worth $697 a year that we can count, and more that we cannot count yet.
6. Material cost never lands against the job, so you are guessing what each job made. We put $0 on this one, deliberately, and Section 3 explains why that is the honest number.

What matters most, and why it is first. Finding 1 is first and it is not because it is the biggest number. It is first because fixing it means switching on a tool that is already inside the Housecall Pro plan you pay for every month. Finding 3 is a bigger single number, but it needs new money and a decision you have already looked at once and said no to. Start where the money is already spent.

The one thing you asked that we want to answer directly. You wrote: how much of all that is the software’s fault and how much is us not using it right, I honestly couldn’t tell you. Every finding in Section 3 ends with that answer in one word: setting, limit, or habit, with a dated source behind it. Two of the six turned out to be the software’s fault in a way you could not have worked out from the outside, and one of those is the payroll problem you have been living with since the summer.

What it adds up to. Between 3.3 and 8.7 hours a week, most of them yours, and $25,171 a year of value with the arithmetic shown. One fix costs money, $5,388 a year, and one fix saves money, $360 a year, so the net is $20,143 a year. Even counting only what we’re certain of: $6,840/yr.

You asked what you would do with ten hours a week back. The honest answer is that we found between three and nine, not ten, and about two thirds of it is your own time rather than Elena’s.

One more thing, and it is yours rather than ours. When you added up your software invoices for us, you found you are spending $482.50 a month, which is $5,790 a year, and you told us you had never added it up before. That was your arithmetic, not ours. Section 3 finding 2 shows where $360 a year of it comes back for no effort at all, and Section 7 shows what each line costs you today against what the vendor charges a new customer now.

What happens next. Section 6 is a week by week plan you could start on Monday. Your two weeks of questions start the day this arrives: reply to the delivery email and we answer. If you think a number is wrong, tell us which one and we will recompute and send you a corrected copy.

2. What we looked at, and how

What you gave us. Your intake form, submitted 2026-09-01, all 33 answers across the six blocks: the business and its shape, the tools you run, how a job travels from first contact to paid invoice, where things stall, what eats your week and your staff’s week, what you have already tried with AI, what you do not want automated, and your rough volumes.

You also offered read-only access to two systems. We did not take it, and you should know that, because it changes how much weight the numbers below can carry. Instead we asked you six questions by email and Elena pulled the answers out of Housecall Pro, the Quotes Awaiting Answer sheet and your invoices. Every customer figure in this report is a number you gave us, and it is labeled that way wherever it appears. We did not log into anything.

What we counted, and where each count came from.

Anything below that did not come from that table is an assumption, and Section 7 lists every one of them.

As is: how a job travels from first contact to paid invoice.

A customer calls or texts, usually about a leak, a clogged drain or a water heater. If Elena is at her desk she answers, takes the basics and books it in Housecall Pro, normally same day or next day. A tech goes out and diagnoses it. If it is straightforward, he quotes on the spot from his phone and does the work on the same visit. If it is bigger, a repipe or a sewer line, he photographs it, you look at it that night or the next morning, and you call the customer with a number. If they say yes, Elena schedules it. When the work is done the tech closes the job in Housecall Pro, which is supposed to raise the invoice automatically, and the customer pays by card in the app, or you post an invoice on net 30 for the commercial accounts. Every week or two, Elena checks that what is in Housecall Pro actually landed correctly in QuickBooks.

Where data gets retyped between systems.

• Invoice and payment detail between Housecall Pro and QuickBooks, when the sync does not carry it cleanly. This is finding 2.
• Technician hours, when the payroll timesheet route breaks and Dana works off a printed report instead. This is finding 4.
• Material costs, which are ordered from the supply house and never land against the job at all, so per-job profit is estimated rather than calculated. This is finding 6.

Where jobs wait or fall through.

• Between “I called them with a number” and “they said yes” on a big quote. Nobody owns the chase. This is finding 1.
• After hours, when a call reaches voicemail and the caller does not wait until morning. This is finding 3.

Process inventory: the twelve manual workflows that run more than weekly

Hours are per week and are recoverable-before-caps, meaning the time the task takes today, not the time a fix would give back. Where a row says “not quantified”, we mean it: the intake told us the work happens but not how long it takes, and we would rather leave the cell empty than fill it with a guess.

Four of twelve rows carry hours. Eight do not, and here is the honest reason for each. Rows 1, 3, 7 and 8 are inherent work: answering the phone and quoting a job is the business, not an overhead to be removed, so their hours would not belong in a recovery total even if we had them. Row 2 is counted, but in finding 3 rather than here: the morning callbacks are the after-hours problem, and timing them in both places would count the same hours twice. Rows 4, 11 and 12 are activities your intake describes but does not time, and none of the six questions we sent you asked for their duration. If you want them counted, tell us during your question window and we will recompute at no charge.

The total in that table is larger than the total in Section 5, and it should be. This table is the time the work takes now. Section 5 counts only the portion a fix can realistically remove, after the caps in Section 7 are applied.

3. The six findings, in payback order

Ranked by return per hour of your effort, not by size. Every finding ends with the same one-word answer to your own question: is this a setting somebody never switched on, a limit in the product itself, or a habit the tools would happily support you changing.

Finding 1: Big quotes go quiet, and nobody owns the chase

Worth $11,613 to $23,225 a year. Confidence: High on the labor, Medium on the recaptured work.

What it is. Elena keeps a spreadsheet called Quotes Awaiting Answer. That spreadsheet is your actual sales record, and you told us so yourself: it “lives entirely in her head and one spreadsheet”. You named chasing quotes as the single biggest thing eating your week, Elena named tracking them as the biggest thing eating hers, and you named it again as the place work stalls. Three of your own answers point at the same object.

What it costs you now.

The labor of doing it by hand:
  you    1.0 to 2.0 hr/wk x $51/hr x 50 weeks = $2,550 to $5,100
  Elena  0.5 to 1.0 hr/wk x $29/hr x 50 weeks = $  725 to $1,450
                                       total  = $3,275 to $6,550/yr
  recoverable at 80%                          = $2,620 to $5,240/yr

The work that never closes because nobody followed up, from your own sheet:
  July     46 quotes, 14 big, $79,300 quoted, 9 closed (64%) worth $48,900
  August   43 quotes, 12 big, $68,750 quoted, 7 closed (58%) worth $39,200
  two months: 26 big quotes out, 16 closed, 10 unanswered worth $59,950
  over a year: about 60 unanswered big quotes worth $359,700, averaging $5,995 each
  at 5% recaptured:  3 more jobs = $17,985 of work x 50% gross margin = $8,993/yr
  at 10% recaptured: 6 more jobs = $35,970 of work x 50% gross margin = $17,985/yr

We are allowed to claim six times more than we have, and we are not going to. The rules we work to would let us model recapturing 30% of unanswered quotes, which would be 18 jobs and $107,910 of work. We modeled 5%. The reason is in your own numbers: your close rate on big jobs is already about 60%, and that already includes the late yeses that come in after you chase somebody. The easy recaptures are inside your existing rate. What a follow-up schedule adds is only the ones lost to silence rather than to a decision. Three extra big jobs a year, one every four months, is what we will defend.

The fix, and it costs nothing. Housecall Pro’s Sales Proposal Tool is included in your Max plan. Their pricing page lists it as a “$40/mo value” and charges $40 a month for it on the Basic and Essentials plans; on Max it is bundled and you are already paying for it. It sends automated follow-up reminders on proposals that have had no response. That is the entire job the spreadsheet is doing badly (housecallpro.com/pricing and housecallpro.com/features/sales-proposal-tool/, both read 2026-09-01).

Setup is about two hours: turn the tool on, rebuild the open items from Elena’s sheet inside it, and set a follow-up cadence. We would suggest day 2, day 7 and day 14, then a call from you.

What it will not do, so you are not surprised. Estimates do not push across to QuickBooks (help.housecallpro.com/en/articles/6293215, read 2026-09-01), so your accounts will still not show a forward view of quoted work. If you want that, it stays a manual report for now.

One boundary, in your words. You told us you do not want a system quoting a repipe automatically, and you do not want anything that sounds like it is not really you talking to a customer who has used you for ten years. Nothing here touches either. The price stays yours. What changes is that the reminder to follow up stops depending on you remembering.

Verdict: habit, sitting on a setting nobody switched on. The tool has been in your plan the whole time. The spreadsheet exists because the tool was never turned on, and the habit grew around the gap.

Finding 2: You check Housecall Pro against QuickBooks by eye

Worth $4,220 to $13,100 a year. Confidence: High.

What it is. Every week or two, you and Elena compare what Housecall Pro says happened against what QuickBooks recorded. On your own description this ranges from “an hour” to “the whole Saturday”. At 60 jobs a week you are eyeballing something like 3,000 invoices a year.

What it costs you now.

  Elena  1.0 to 2.5 hr/wk x $29/hr x 50 weeks = $1,450 to $ 3,625
  you    1.5 to 5.0 hr/wk x $51/hr x 50 weeks = $3,825 to $12,750
                                       total  = $5,275 to $16,375/yr
  recoverable at 80%                          = $4,220 to $13,100/yr

Why only 80%, and this one is not a hedge. Some of this checking can never go away, because of what the connection between the two systems actually is. Housecall Pro’s own documentation says the sync is one way, from Housecall Pro into QuickBooks, and that it carries exactly four things: invoices, payments including those on Recurring Service Plans, new customer records, and new Price Book items. It also says, in as many words, that “Estimates do not push over to QuickBooks Online”, that invoices you delete in one system must be deleted by hand in the other, that changes to a job do not push unless you use an action button, and that manual edits to Price Book items do not sync at all (help.housecallpro.com/en/articles/6293215, read 2026-09-01).

So estimates, deletions and job edits will always need a human. That is the 20% we did not count.

The fix. Two hours, free, and in three parts. First, switch the sync trigger to fire when a job is marked finished rather than leaving it to a manual push, so the gap closes at source. Second, agree a short written list of the four object types above and stop checking anything outside it, because those are the only things the connection ever promised to carry. Third, replace the by-eye comparison with one monthly report run against those four types. Most of what you are checking today is work the connection was never doing in the first place.

Verdict: mostly a setting, with a real limit underneath it. The bulk of it is configuration and scope you can fix this week. The residue is a genuine product limit and no amount of setup removes it.

Finding 3: After-hours callers reach voicemail, and some hire somebody else by morning

Worth $8,438 to $16,031 a year. Confidence: Medium.

What it is. You called this the task everyone hates most: “nobody likes starting the day apologizing to five people for not picking up the phone the night before, and by then a couple of them have already called somebody else.” Elena pulled two weeks of the call log for us and it turns out you were describing it accurately.

What it costs you now.

  week of Aug 10: 14 after-hours calls, 9 voicemails, 5 hangups
  week of Aug 17: 11 after-hours calls, 7 voicemails, 4 hangups
  of the voicemails, 3 of 9 and then 2 of 7 had already hired somebody by the callback

  confirmed lost: (3 + 2) / 2 weeks = 2.5 callers/wk x 50 = 125 a year
  at 30% recaptured: 37.5 jobs x $450 = $16,875 of work x 50% margin = $8,438/yr
  upper end, counting half the hangups as lost too:
    4.75/wk x 50 = 237.5 x 30% = 71.25 jobs x $450 = $32,063 x 50% = $16,031/yr

We show the value of the work and the margin on it, because $450 of recovered revenue is not $450 of value to you. The margin is what you keep. We used your two weeks of log rather than the 30 calls a day from your intake form, because that figure is your total call volume and not the after-hours part of it.

Now the part that matters more than the arithmetic. You already looked at this. You told us Elena researched an AI phone answering service last year, watched a demo, and you never switched it on, because it “felt like it was built for a call center, not for a plumber who might be under a sink”, and because you were not confident it would book the job correctly or know a real emergency from a routine call. That is a good objection and most of the market deserves it. So here is what we checked, by name, and what we found. Everything below was read on 2026-09-01.

The detail behind the first three verdicts, since the table only has room for the answer. Sameday AI’s plumbing page says it “qualifies burst pipes, active leaks, and no-water calls, applies your emergency criteria, and transfers true emergencies to your on-call plumber while scheduling everything else automatically”. Your criteria, not theirs, which is the exact thing you said was missing. Avoca AI’s published positioning aims at operators above $3M in revenue with five or more office staff taking 50+ calls a week; you have Elena. They also have no pricing page, and we will not repeat the figures other websites quote for them, because those are not Avoca’s numbers. Housecall Pro’s CSR AI page talks about setting “clear booking and escalations rules”, which sounds right, but it does not document letting you define what counts as an emergency, and that is your whole question; other write-ups describe it spotting words like “emergency” and “urgent”, which is not the same as applying your rules.

Two things we are telling you not to waste an afternoon on: Answering Legal and Rilla are both category errors that show up in searches for this problem. Neither one answers a plumbing call.

A cap you should ask about before you sign anything. Sameday’s $449 tier includes 500 voice minutes a month and their pricing page does not publish what a minute costs after that. Your after-hours volume is small, roughly 12 calls a week, so 500 minutes is probably comfortable for after-hours only. It is almost certainly not enough if you ever point the whole 30-calls-a-day at it. Get the overage rate in writing before you sign. We have not modeled an overage, because we cannot source one.

The budget tension, said plainly rather than smoothed over. You told us you are comfortable spending $250 to $500 a month on new tools. At $449, Sameday fits inside that. But you are already spending $482.50 a month on software, so saying yes takes you to about $931 a month, which is close to double. That is a real decision and it is yours. Three honest options:

1. Do nothing. This is a legitimate answer and we are not going to pretend otherwise. The finding is Medium confidence, it rests on two weeks of log, and $449 a month is a lot next to a $8,438 estimate. If you do nothing here, the other five findings still return about $16,733 a year at the low end, and every one of them is free.
2. Measure first, then decide. Spend nothing. Elena can keep a tally for four more weeks of after-hours calls and how many had already hired someone by the morning. If 2.5 a week holds up, the case is much stronger than two weeks can make it. This is what we would do.
3. Buy it, but start narrow. After hours and weekends only, not daytime overflow. Get the overage rate first. Write your emergency rules down before the demo: burst pipe, active leak and no water go to the on-call phone; everything else books for the morning. Then judge the demo against your own list rather than against theirs.

Section 6 puts this decision in week 4 on purpose, after the free fixes have started returning roughly what it costs.

Verdict: neither a setting nor a habit. This one needs something you do not currently own, which is precisely why it is third on the list and not first.

Finding 4: Payroll gets redone by hand, and not for the reason you think

Worth $203 to $813 a year. Confidence: Medium.

What it is. You pay for the QuickBooks Payroll add-on. It is supposed to take the hours from Housecall Pro timesheets. It does not, often enough that Dana works off a printed report instead. Your bookkeeper’s count: over the last four months, 8 runs, clean on 3, done by hand on 5. You told us it seemed to get worse after a Housecall Pro update in the summer.

What it costs you now.

  biweekly runs                            = 26 a year
  manual on 5 of 8                         = 62.5%, so about 16.25 runs a year
  16.25 x 0.5 to 2.0 hr x $50/hr (Dana)    = $406 to $1,625/yr
  recoverable at 50%                       = $203 to $813/yr

This is the smallest number in the report and possibly the most useful sentence in it. We went looking for the broken integration so we could tell you how to fix it. We could not find it, because as far as we can establish it does not exist and never has.

Housecall Pro’s Time Tracking feeds Housecall Pro’s own payroll product, at its Advanced tiers (housecallpro.com/features/payroll/ and help.housecallpro.com/en/articles/8594391). Separately, Housecall Pro’s QuickBooks integration carries invoices, payments, new customers and new Price Book items, and nothing else (help.housecallpro.com/en/articles/6293215). Timesheets are not on that list. We read six of their own surfaces and none of them documents hours crossing from Housecall Pro into Intuit’s payroll product on any plan, at any price. All three read 2026-09-01.

We want to be precise about how strong that claim is. We did not find a vendor statement saying “this is not supported”. We found consistent absence across every surface where it would be documented if it existed. That is weaker than a denial and much stronger than a guess, and you should weigh it that way. If your Housecall Pro rep tells you otherwise, ask them for the help article, and tell us, because we would want to correct this.

So nothing broke in the summer. What most likely happened is that a change made a manual workaround stop working, and it had always been manual underneath.

Our recommendation is to keep doing it by hand. That is a real recommendation and not a shrug. The problem is worth about $400 a year at the midpoint. Every route to automating it costs more than that in money or in risk:

• Move to Housecall Pro Payroll (their Advanced tiers). The timesheet link is genuine, but you would be moving payroll systems to save a few hundred dollars, and your accounts would then span two vendors differently than they do now.
• Move to Gusto, which has well-documented time tracking and payroll together. Same objection: the switching cost dwarfs the problem.
• Keep re-keying, and stop waiting. Dana already has a working method. The only thing you gain from this finding is the end of the expectation that it will start working by itself, and the ten minutes a fortnight you spend investigating it.

Verdict: a limit. Not your fault, not a setting, and not something a support ticket will fix.

Finding 5: Comfort Club members lapse with nothing chasing them

Worth $697 a year that we can count. Confidence: Medium.

What it is. You have 145 Comfort Club members at $199 a year. Over the last twelve months 38 joined and 24 lapsed. Nothing automatic asks a lapsing member to renew.

What it costs you now.

  24 lapsed x $199                     = $4,776/yr of recurring revenue walking out
  at 30% recaptured                    = 7.2 memberships, which we round DOWN to 7
  7 x $199 = $1,393 of work x 50% margin = $697/yr

What we deliberately did not count, and it is the bigger number. You estimated that about one in three membership visits turns up $300 to $350 of extra paid work. At 145 members that would be a five-figure line in this report. You told us plainly that it was a guess with no system behind it, and Elena offered to pull real figures if it mattered. We left it at zero. A five-figure finding resting on an estimate is how audits get written that fall apart the first time somebody checks them.

If you want it counted, that is exactly what your two-week question window is for. What we would need is twelve months of membership visits and which of them produced a second invoice. Elena can pull it, and we will recompute at no charge.

The fix, and it costs nothing. Recurring Service Plans is a Max-plan feature and you are on Max (housecallpro.com/pricing, read 2026-09-01). It holds memberships, bills them, and schedules the visits. Your members are currently in a spreadsheet. Moving 145 records across is about three hours of Elena’s time, most of it a one-off import, and after that the renewal chase happens without anybody remembering to do it.

Verdict: a setting you are already paying for and have not switched on.

Finding 6: Material cost never lands against the job

Worth $0 in this report, on purpose. Confidence: Medium.

What it is. You order from the supply house and, in your words, you “still eyeball job profitability instead of tracking it properly per job in the system”. So you do not know which jobs make money. You suspect some do not.

Why the number is zero. To size this we would need the per-job material cost, and that is exactly the figure you do not have. We could construct something from your $450 average and a typical material share, but it would be our assumption multiplied by your average, presented as your finding. That is padding, and we would rather hand you a zero we can defend. What this finding is worth is the decisions it changes once you can see it, and that is real even though we are not putting a dollar on it.

This is where the report saves you money by telling you not to buy something.

Here is the trap. The obvious move is to upgrade QuickBooks so it can do job costing. You have QuickBooks Online Essentials, and it cannot: Projects is a Plus and Advanced feature. Plus is $140 a month against Essentials at $85, so the upgrade is $55 a month, $660 a year (Intuit plan pages, read 2026-09-01).

Do not spend it, for two separate reasons.

1. It would not work anyway. The Housecall Pro connection carries invoices, payments, customers and Price Book items. It does not carry cost data at all (help.housecallpro.com/en/articles/6293215, read 2026-09-01). You would be buying a job costing feature and then hand-feeding it, which is the problem you already have.
2. You already own the answer. Housecall Pro has Job Costing built in, available from the Basic plan upward, so it is in your Max plan today. It tracks labor, materials, commissions and miscellaneous expense against the job, and it pulls unit cost straight from your Price Book automatically (help.housecallpro.com/en/articles/6536636, updated 2026-07-22, read 2026-09-01, and housecallpro.com/features/job-costing/).

The fix. The work is not switching it on, it is putting real unit costs into your Price Book, and that is a genuine job: perhaps four to six hours with your supply house invoices open. Do the twenty items you fit most often and leave the tail. After a quarter of jobs running through it you will have the real number, and this finding stops being a zero.

Verdict: a limit in one product and a setting in the other. QuickBooks Online Essentials genuinely cannot do it. Housecall Pro genuinely can, and nobody switched it on.

4. What we deliberately did not pursue

Everything we looked at and left, with the reason, so you can see it was decided rather than missed.

• Moving off Housecall Pro. Nothing in your intake asks for it, four of the six findings above are fixed by features it already has, and a migration would cost more than every finding in this report combined.
• Moving off QuickBooks Online. Same reasoning. Your $67.50 rate is below current list, so a move would cost you money on day one.
• The QuickBooks Online Plus upgrade for job costing. Ruled out inside finding 6, at $660 a year, because Housecall Pro already does it and the connection would not carry the cost data anyway.
• Automating the price on a repipe or a big repair. Your standing rule, stated in your intake, and we agree with it. We looked only at how fast a price reaches the customer, never at who decides it.
• Anything that sounds like it is not really you texting or calling a ten-year customer. Your words, and they bind the fixes in findings 1 and 3.
• The truck stock and on-call rota spreadsheet. You told us you do not trust the app’s inventory tracking. The sheet works, the payback is small next to findings 1 to 4, and we are not going to spend your first free hours on it.
• The Facebook page you barely post to. No finding here would clear the bar we hold ourselves to.
• Google Local Services Ads and your Google Business Profile. Both are working and neither showed up in your answers as a stall or a time sink. Bonus A covers the part of this worth your attention.
• Anything above $500 a month. Your stated ceiling for new tools, and it ruled out several otherwise reasonable options before we ever wrote them down.
• The time it takes to quote a big job (row 4 of the Section 2 inventory). Your intake describes it but does not time it, and we did not ask. It may be a finding. We are not going to invent it into one.

One of your named pains has no finding, and you should know which. You listed reviewing invoicing and the books on Fridays and Saturdays as the third thing eating your week. That is finding 2 and it is counted there. The rest of that time is inherent work: somebody has to look at the books.

5. What this is worth, and the bar we hold ourselves to

                                          low             high
  1. Big quotes go quiet              $11,613         $23,225
  2. Housecall Pro vs QuickBooks      $ 4,220         $13,100
  3. After-hours callers              $ 8,438         $16,031
  4. Payroll by hand                  $   203         $   813
  5. Comfort Club lapses              $   697         $   697
  6. Material cost per job            $     0         $     0
                                      -------         -------
  total identified                    $25,171         $53,866
  new tool spend (Sameday AI)         -$5,388         -$5,388
  saving found (annual billing)         +$360           +$360
                                      -------         -------
  net                                 $20,143         $48,838

• Total identified opportunity: between 3.3 and 8.7 hours a week and $25,171 a year.
• New tool spend: $5,388 a year, all of it finding 3, and only if you decide to take it.
• Saving found: $360 a year, finding 2’s annual-billing switch, which costs you nothing.
• Net: $20,143 a year.
• Even counting only what we’re certain of: $6,840/yr. That is findings 1 and 2, the two rated High, counting only labor we can remove and no revenue we hope to win back.

The bar we hold ourselves to. We do not send an audit that finds less than $4,500 a year against a $1,500 fee. This one finds $25,171. That is our own standard for our own work, and it is the reason this report exists in the shape it does rather than a promise we are making to you.

Where we have been deliberately conservative, so you can see the shape of the caution rather than take our word for it:

• Every headline figure above is the low end of its range.
• Finding 1’s recapture is modeled at 5% when the rules we work to would allow 30%. At 30% that line alone would read $53,955 instead of $8,993.
• Finding 6 is $0, and your one-in-three follow-on estimate in finding 5 is excluded entirely.
• Finding 5 rounds 7.2 memberships down to 7.
• The hourly rates are burdened, not fully loaded: they include wages and the employer’s benefit cost, and exclude trucks, tools, insurance, rent and software. That understates every hours-based finding here.
• Your own hourly rate is used at $51 rather than the $65 you estimated, because the sourced market figure came in lower and we take the lower of the two.

6. The first 30 days, in order

Free and fast first. The paid decision comes last, after the free work has started paying for it.

Total cost of weeks 1 to 3: $0. Total return from weeks 1 to 3: about $16,733 a year at the low end, plus the $360 saving. Week 4’s decision costs $5,388 a year and returns $8,438 at the low end.

If you only do three things: switch to annual billing (ten minutes, $360). Turn on the Sales Proposal Tool (two hours, the largest finding in the report). Set the sync trigger and stop checking what the connection never carried (two hours, the highest-confidence finding).

Who decides what. Every “you” above is you, not a vendor. The only step that needs anybody outside your business is week 4, and it points at the priced shortlist in finding 3, not an open search.

7. Appendix: our assumptions

Every judgment call sitting on top of the numbers, so that disagreeing with one rescales the rest instead of discrediting the report.

The three hourly rates, and both figures for each

Per your instruction we show what you told us alongside what we sourced, and we use the lower.

Why each of those three is what it is. Your $51 is the Texas 75th percentile for plumbers, pipefitters and steamfitters (SOC 47-2152), $35.63/hr, times the employer benefit load below. It is lower than your own estimate, so we used ours. Elena’s $29 is your own figure: Texas secretaries and administrative assistants (SOC 43-6014) run $31.01 to $36.78/hr burdened, median to 75th percentile, and yours is lower, so we used yours. Dana’s $50 is her contracted rate, which is a real invoice and not an estimate; a contractor’s billed rate already includes their costs, so no load is added, and the cross-check is Texas bookkeeping clerks (SOC 43-3031) at the 90th percentile, burdened.

Wage source: US Bureau of Labor Statistics, Occupational Employment and Wage Statistics, 2025 reference period, Texas statewide. Read 2026-09-01 through O*NET OnLine’s state wage tables, which republish the OEWS state file; bls.gov itself refused our requests that day.

Benefit load: US Bureau of Labor Statistics, Employer Costs for Employee Compensation, March 2026, released 2026-06-12. Private industry total compensation $46.60 per hour worked against wages and salaries of $32.60, which is a multiplier of 1.429.

Three honest limits on those rates. First, they are burdened, not fully loaded: wages plus the employer’s benefit cost, with no truck, tool, insurance, rent or software overhead, because no free and checkable overhead benchmark for residential plumbing was available to us. That understates every hours-based finding. Second, the wage survey excludes the self-employed by definition, and you are an owner-operator, so your rate is drawn from employed master plumbers as the closest defensible proxy. Third, we wanted the benefit load for small goods-producing employers specifically, which would be lower than the all-industry figure we used; that table was unreachable on the day, so we used the higher published figure and then took your wage base at the 75th percentile rather than higher up the band, to avoid banking an increase we could not source.

Volume and working-week assumptions

• A 50-week year throughout, which allows two weeks off. If you work 52, every annual figure here is about 4% low.
• Given by you: 60 jobs a week, 30 calls a day, 45 open quotes a month, $450 average job, 145 members, the two weeks of call log, the July and August quote figures, the payroll counts, the tool invoices, the membership joins and lapses.
• Estimated by us: the hours-per-week ranges in findings 1 and 2. They come from your own descriptions, including “some weeks that’s an hour, some weeks it eats the whole Saturday”, and they are ranges rather than points because that is genuinely how well we know them.
• Gross margin: 50%, used wherever we convert recovered work into value. Published benchmarks for residential plumbing service and repair run 50% to 65%; whole-trade figures that include new construction run lower. Publishers disagree mainly on whether the truck is a cost of the job or an overhead, which is worth ten points or more by itself. We took the floor of the service and repair range.

How much of each finding we counted

We cap what we claim according to how confident we are, and we never claim the whole of anything.

Methodology

We read your intake, drafted a findings list, had it reviewed before any prose was written, then asked you six questions to fill the gaps that mattered most. Every capability claim in Section 3 was checked against the vendor’s own current documentation and carries the date we read it, because prices and features move monthly. Where a vendor does not publish a price we say so rather than repeating a third party’s estimate: that applies to Housecall Pro’s CSR AI add-on, its Pipeline add-on, and Avoca AI.

Where a claim rests on absence rather than a statement, we say that too. Finding 4 is the case: we established that no Housecall Pro to QuickBooks payroll timesheet route is documented on any of the six vendor surfaces where it would appear, which is not the same as the vendor saying it is unsupported.

We had no access to your systems. Every figure attributed to you in this report is a figure you or Elena sent us. We did not log into Housecall Pro or QuickBooks, and nothing here was verified against a live account.

Delivery timeline

Data handling

This is a fictional sample, so there is no customer data to hold or delete. On a real order we work from what you send us plus any read-only access you grant, we delete shared documents and revoke access within 30 days of delivery, and we keep the report and the numbers worksheet as work product under the agreement.

How to get help

Two weeks of email questions, starting the day this arrives. A check-in from us on day 10. A live 30-minute readout call, included with your audit, booked through the link in your delivery email.

Bonus A: Getting named when a customer asks an AI assistant

More of your customers are starting a search by asking an assistant rather than typing into a search box. “Who is a good plumber near me for a burst pipe at 9pm” is a question people now ask a chat window, and the answer they get is assembled from what those systems can find and trust about local businesses.

Nobody can guarantee placement in an AI answer. Anyone who tells you they can is selling you something. The systems change their behavior without notice, they differ from each other, and the same question asked twice can produce different answers. What you can do is make the fundamentals findable and correct, which is the same work that makes you findable to a human, and then re-check it from time to time. That is the whole of the honest version of this.

What we would normally report here, and why this section is short. For a real customer we run a checklist against the business’s own public footprint. Cobble Creek Plumbing and Drain does not exist, so there is nothing to check and no assistant could name it. Rather than invent a result, this section gives you the checklist itself, which is the part that transfers.

The input checks, and what each one is for.

What to do about a gap, in the same three shapes as the rest of this report. Do it yourself, most of these are an afternoon each. Hire anyone, a local web person can do the structured data and the directory work for a few hundred dollars. Or we can do it as a fixed-price piece of work. None of it routes only through us.

This section is descriptive and it is not in any total in Section 5. There is no reliable published figure for what an assistant mention is worth, so we do not put a dollar on it and it does not enter the value bar. The engine list, the crawler names and the products in this space all change quickly, so treat this checklist as current at 2026-09-01 and worth re-reading in six months.

Bonus B: A starter AI policy

The most common worry owners raise about AI is not cost, it is what happens to their customers’ data. This is a starting point you can adapt. It is not legal advice, and it is worth twenty minutes of your lawyer’s time before you put it in front of staff.

1. Approved tools.

2. What never goes into a general AI tool. Customer names paired with addresses. Phone numbers. Card or bank detail. Photographs of the inside of a customer’s home that show identifying features. Employee personal data. Anything from a commercial contract you have signed.

3. Customer data. If a tool is not one you pay for and control, treat anything typed into it as public. Strip names and addresses before pasting job notes. “A two-story house with a leak under the kitchen sink” is fine; the same sentence with the address is not.

4. Acceptable uses. Drafting and rewriting text you then check. Summarizing a long document. Explaining an error message. Working out how to word something awkward to a customer. In every case a person reads it before it goes out.

5. Prohibited uses. Setting a price on a job. Sending anything to a customer that a person has not read first, which is your own standing rule. Making a hiring or discipline decision. Passing off AI output as if a person wrote it in a situation where that matters. Uploading a signed contract.

6. If something goes wrong. Report it to Marcus the same day, and write down what was entered and into which tool. If customer data was involved, note whose. Do not delete anything before it has been recorded.

7. Review. Read this again every six months, and whenever you adopt a new tool. The tools change faster than the policy does.

AI Operations Audit: Sample

Illustrative sample, fictional business

Prepared for Bramblewood Bookkeeping and Advisory, LLC · Tulsa, Oklahoma · 2026-09-01

Prepared by Iterategy · SAMPLE-BRAMBLEWOOD · Turnaround: 7 business days

1. Read this first

This sample audit describes a business that does not exist. Bramblewood Bookkeeping and Advisory, its owner Renata Kessler, and her bookkeeper Deshawn are invented, and every number below is fabricated but internally consistent. It is published to show what the real thing looks like. No real client is referenced anywhere in it.

What this is. You bought a written report that reads your intake answers, counts the hours in your week, and puts a dollar figure and the arithmetic behind every finding. This section is the whole audit in plain language. It takes about five minutes and it stands on its own.

What we found. Six things, in the order they pay you back.

1. Chasing clients for documents is not unfixable. It is unautomated. You told us it is the task everyone hates most and that you do not think it is fixable. It is worth $4,900 to $9,850 a year, and you are partly right about why.
2. The line between what a rule can do and what your judgment has to do. Categorization and reconciliation is the biggest block of hours in the firm, yours and Deshawn’s together. Two thirds of the transactions on your messiest file are vendors that were there last month. Worth $4,395 to $9,390 a year.
3. Thirty-eight close summaries written from a blank page every month. Worth $2,775 to $6,300 a year, and part of that gain you have already banked yourself.
4. New client details typed into three systems. Worth about $102 a year. We are counting it at zero and telling you to leave it alone. You were right.
5. What you actually pay for software. You said a couple hundred a month. It is $417. There is no saving in that sentence, but there is something you should see.
6. The three or four more clients you asked about. We are counting this at zero dollars and answering it in hours instead, because hours are the thing we can check.

What matters most, and why it is first. Start with the document chasing, not with the bigger number underneath it. Not because it is worth more, but because the first move takes about twenty minutes, costs nothing, and uses a feature you are already paying for and have switched on for four clients out of fifteen. The categorization finding is worth about the same and takes a couple of evenings of setup. Do the cheap one first.

What it adds up to. Between 7.0 and 13.4 hours a week, worth $12,070 to $25,540 a year. Even counting only what we’re certain of: $7,170/yr.

Now the part you should read twice. You asked for ten hours a week. The low end of what we found is 7.0 hours, and only 4.8 of those are yours. The rest are Deshawn’s, and Deshawn’s hours cannot review a new client’s books. So the honest answer to your question is that the low end does not reach ten, and the top end comfortably does, and where you land inside that range depends almost entirely on how far you take the second finding.

And the Saturdays. You said you would take some of the time back for yourself and stop working Saturday mornings in tax season, and put the rest toward three or four more clients. Those two things draw on the same recovered hours. Averaged over the year the low end pays for both, with about sixty hours to spare. Inside tax season it does not. Your twelve Saturdays are 48 hours falling in twelve consecutive weeks, and those are the same weeks your volume peaks and Deshawn is already at thirty hours. So it is a real choice, not a free one, and you should make it deliberately rather than discover it in February.

What happens next. Section 6 is a thirty-day plan in the order we would do it, free and fast first. Your questions are open for two weeks from today: reply to the email this came with and we answer. If you find a factual error, tell us and a corrected copy is back to you inside two business days at no charge.

2. What we looked at, and how

What you gave us. Your intake form, submitted 2026-09-01, all thirty-four answers. Then a follow-up set of twenty-six questions, because the form named every task in your week and quantified none of them. Your answers to those questions are the backbone of this report, and everywhere a number came from you rather than from a system we have said so in the words “you told us”.

What you gave us access to. You granted read-only access to your QuickBooks Online Accountant view and your Financial Cents admin view. Both are your own screens rather than any client’s books, and that distinction runs through every finding in this report. The transaction counts below came out of the QuickBooks Accountant view. Nothing in this audit opened a client file.

What we counted, and how you can re-run it.

How a month moves through your firm today. A client signs after a twenty-minute Calendly call and a PandaDoc engagement letter. Their details get typed into PandaDoc, then again into QuickBooks to open the company file, then again into your Client Tracker sheet. Then the monthly cycle starts and it is the same for all thirty-eight: transactions categorized weekly or every other week, reconciliation at month end, anything unclear chased down, and a close summary email out by the tenth of the following month. Invoicing runs on autopay through QuickBooks Payments. In your words, that part runs on rails.

Where it stops running on rails. Two places, and you named both.

• Waiting on the client. A missing bank statement, an unanswered question about a $4,200 charge, a receipt nobody uploaded. This is what pushes closes past the tenth.
• The priority list. Categorization sits in a queue longer than it should because you and Deshawn are not always working from the same one.

Hand re-entry points. Three that you named, and two more that Deshawn added when we asked him: new client details across PandaDoc, QuickBooks and the Client Tracker; client W-9s and vendor details typed into QuickBooks and then again into Gusto for the nine payroll clients; and the monthly Dext export matched by hand to QuickBooks on the four clients whose bank feeds lag.

The ten most-repeated manual workflows

That 43 hours a week across 38 clients is about 1.13 hours per client per week, and it is the number Finding 3.6 is built on.

One thing we could not count, and are not pretending we did. Row 9, the Dext-to-QuickBooks matching Deshawn named, has no hours figure because nobody has ever timed it. It is not in any total in this report. Tell us what it runs to and we will recompute at no charge inside your question window.

3. The six findings, in payback order

Ranked by what you get back per hour of your own effort, not by size. The twenty-minute fix that returns $4,900 comes before the two-evening one that returns about the same.

A note that applies to all six. You told us your rule: nothing goes to a client without your review, because your name and your license are on the work. Every fix below is built that way. None of them sends a client anything you have not seen, and none of them gives any new company access to any client’s books. That last point is not an accident; it is a constraint we set before writing a word, and two otherwise attractive ideas were cut by it. They are in Section 4.

3.1 Chasing clients for documents

Your most hated task, and the one you told us cannot be fixed.

What you told us. “Chasing late clients for bank statements and receipts every single month. It’s the same conversation on repeat and it’s the reason our closes slip past the 10th.” And then, about the same task: “that one I don’t think is fixable, it’s just what clients are like.”

You are partly right, and the finding has to concede it before it argues. A reminder does not make a client answer the question about the $4,200 charge. Nothing in this report makes a disorganized client organized. What a reminder system removes is not the client’s slowness. It removes you being the reminder. If this finding wins, it wins on that distinction and on nothing else.

Current cost.

you       6 to 8 hr/wk x 50 weeks x $37 to $42/hr  =  $11,100 to $16,800
Deshawn   2 hr/wk      x 50 weeks x $29/hr         =  $ 2,900
                                            total  =  $14,000 to $19,700 a year

What is recoverable. 35% at the low end, 50% at the high end. $4,900 to $9,850 a year, and 2.8 to 5.0 hours a week, of which 2.1 to 4.0 are yours.

Why not more than 50%. Two reasons, both from your own answers. First, you told us the split of what you chase: 40% bank and card statements, 35% receipts, and 25% questions about what a charge was for. That last quarter is not a document request. It is a question only the client can answer and only you can ask well, and no system touches it. Second, this finding is Medium confidence, so 50% is its ceiling and we have not gone near it.

What does it today. You, from your own inbox, one client at a time, from memory of who owes what. Deshawn sends the first round on his own files and it comes back to you when it is the second or third ask.

What could. Three things you already pay for, in the order they cost you least.

1. Dext, Request paperwork, on the other 11 of your 15 Dext clients. It sends the client a notification in the Dext app naming the specific missing receipt, and it carries an in-app chat for document queries. You have it on for 4 of 15 because you were worried it would annoy people. You have now had it running on four clients for long enough to know whether it does. Cost: $0. Time: about twenty minutes. This is the first thing to do in this whole report.
2. Dext, Fetch, on the clients whose feeds break. Fetch connects to a supplier’s portal and collects the bills itself, checking each connected supplier weekly. Statements are 40% of what you chase and you told us it is mostly clients whose bank feeds break or who have an account that is not connected. This is the category that stops being a conversation at all. Cost: $0 on your tier. Time: one client at a time, about ten minutes each.
3. Financial Cents client tasks and requests, for the rest. Your Team plan carries the client portal, client tasks and requests, and secure file sharing. That gives you one place where a request lives with a due date instead of a sent email you have to remember. Cost: $0.

Start with the nine. You told us 9 clients are late every single month, 12 occasionally, and 17 never. Do not build a system for 38. Switch it on for the 9, watch one close cycle, then decide about the 12.

One correction, and it changes what this costs you. You told us your Financial Cents Team plan “does have client requests and a client portal with automatic reminders”. Financial Cents’ published plan table says the automatic part is not on Team. Team carries the portal, the client tasks and the requests. The automated follow-ups are named “Auto-follow ups for client tasks” and they are the first feature of the Scale plan, at $69 a seat a month against Team’s $49, both on annual billing (checked 2026-09-01). Your invoice of $98 a month for two seats confirms you are on Team annual.

So: check your own plan page before you rely on the automated half. It takes a minute and you have the admin view. If we are right, the arithmetic on the upgrade is:

Scale $69 - Team $49 = $20 a seat a month x 2 seats = $40 a month = $480 a year

We have not counted a single dollar of that upgrade in any total in this report. The $4,900 to $9,850 above is what the free path returns. The upgrade is a decision you make after you have run the free path for a month and know whether the manual send is the part that hurts. The do-nothing option is real here: the free path plus your existing habit of sending the reminders yourself is still better than today, because today the request lives in your head.

The data boundary. Everything above is an outbound request from a system you already own to a client who already uses it. No new company gets access to any client’s books. Your engagement letter already names Dext, so nothing here needs a conversation with anyone.

Verdict: habit. The capability is bought and switched off. Sources: Dext Help Centre on Request paperwork and Fetch, and the Financial Cents published plan table, both checked 2026-09-01. Confidence: medium (it depends on the 9 clients responding to an app notification better than they respond to you, which we cannot know until you try it on a cycle).

3.2 Categorization and reconciliation

The biggest block of hours in the firm, and the one place you have already tried the obvious fix and rejected it.

What you told us. Twelve hours a week for you on the harder files, fifteen of Deshawn’s twenty-five on the easier ones. About 9,500 transactions a month across the book, roughly 250 per client, with the three messiest running 700 to 900 each. On one of those 900-a-month files, about two thirds are vendors that were there last month. The gas stations, the supply house, the phone bill, the same subcontractors. The other third needs a human.

And: about a dozen bank rules exist in total, across 5 or 6 of your 34 QuickBooks files, all set up years ago at onboarding. Deshawn has never created one.

Current cost.

you       12 hr/wk x 50 weeks x $37 to $42/hr  =  $22,200 to $25,200
Deshawn   15 hr/wk x 50 weeks x $29/hr         =  $21,750
                                        total  =  $43,950 to $46,950 a year

What is recoverable. 10% at the low end, 20% at the high end. $4,395 to $9,390 a year, and 2.7 to 5.4 hours a week, of which 1.2 to 2.4 are yours.

Why so low, when the number underneath it is so large. This is the finding most likely to be overclaimed, so here is the ceiling we are allowed and the number we chose. Two thirds of transactions on a messy file recur, so two thirds is the absolute maximum a deterministic rule could ever touch. Our own house limit for a high-confidence labor finding is 80% of the addressable share, which would be 80% of 66.7%, or 53%. At 53% this line would read about $23,400 to $25,000 a year. We are claiming 10% to 20%, between a fifth and a third of what we are permitted. The reason is that the two-thirds figure is your eyeball estimate on one file, and the entire finding would move if it turned out to be half or three quarters on the others. Correct that number and we recompute, free, inside your question window.

Now the objection, because it is the right one.

You turned on Intuit Assist’s auto-categorization for one client as a test, it miscategorized, and you turned it off. When we asked what it actually got wrong you said: it put a contractor’s Home Depot purchases in Office Supplies instead of job materials, and it coded an owner draw as a vendor payment, both without hesitation. The routine things, gas, phone, the same subcontractor every month, it got right. What scared you was that it was confident about the judgment calls it had no business making.

That is not an argument against what we are recommending. It is the argument for it.

A bank rule and an AI suggestion are different kinds of thing, and Intuit’s own documentation says so. A rule is a condition you wrote and a category you chose. It tests the condition and applies the category, identically, every month, and it does not estimate. The AI suggestions in QuickBooks are built the other way: Intuit describes them as based on “your past work and transaction details” and on “info in the full bank description and your transaction history”, which is a description of inference from pattern.

The clearest evidence is a feature Intuit ships in the product. Review Signals are color-coded icons on the bank transactions page that show, in Intuit’s words, “the quality of data behind each QuickBooks suggestion”: whether it rests on a strong pattern in your books, on limited data “worth a second look”, or on very little history that “needs your attention before you post”. Intuit also states plainly that QuickBooks “may not have access to all the necessary information to categorize certain transactions accurately” and “relies on patterns and algorithms to make suggestions, but sometimes, additional context is needed.”

A bank rule has no confidence indicator, because it is not guessing. Your two failures were both “additional context” cases: knowing that this client is a contractor and therefore that Home Depot is job materials, and knowing that this particular transfer is the owner taking money out. You have that context. The suggestion engine does not. You were right to turn it off, and you should leave it off. The recommendation is the other tool.

What could.

1. Write rules for the recurring two thirds on the three messiest files first. Those three carry 700 to 900 transactions a month each. A QuickBooks company file holds up to 2,000 rules and a single rule can carry up to 5 conditions, so nothing about your volume is near a product limit. Cost: $0. Time: an evening for the first file, much less for the next two.
2. Then stop writing them one file at a time. You told us you have always done rules per file and had heard there is an export and import but never tried it. There is. In the file that has the rules: Transactions, then Banking, then Rules, then the arrow beside New rule, then Export rules, which saves a spreadsheet. In the target file: the same menu, then Import rules. Your clients are contractors, small retail and similar firms, so the vendor overlap across files is real: the same fuel, phone and supply-house vendors recur. Cost: $0. This is the difference between the fix being 34 jobs and being about 4.
3. Hand the rules to Deshawn. He has never made one, and he is carrying fifteen hours a week of the easier files, which is exactly where recurring vendors dominate. Half of this finding’s hours are his.

Where the real limit is, and we are not going to pretend otherwise. The other third is the part that needs a human, and after every rule you can write, it still needs one. This finding does not touch it, no tool in this report touches it, and the judgment on the harder files is the thing you are actually paid for. The recoverable share stops where your judgment starts, and that is why this is a 10% to 20% finding rather than a 53% one.

The data boundary. Bank rules live inside your own QuickBooks admin view, in files you already administer as the ProAdvisor of record. Nothing is shared with anyone, nothing new is connected, and no third party sees a transaction. Your engagement letter already names QuickBooks.

Verdict: setting. The capability is inside a subscription you already pay for and it is switched off in 28 of 34 files. Sources: Intuit help on bank rules, on importing existing bank rules, on AI suggestions and on the AI-powered banking page, all checked 2026-09-01. Confidence: high on the labor figures, which come from your own hours and your own transaction counts.

3.3 Thirty-eight monthly close summaries

A discrete, countable, monthly task, and the one place in this firm where drafting-for-review fits your rule exactly.

What you told us. About twenty hours a month across all thirty-eight, so five hours a week. Roughly twenty-five minutes each on average, but not evenly: 8 clients take about half the time, the ones with two entities, a loan, or an owner who asks questions. The other 30 run 10 to 15 minutes each. Separately, about an hour a week of call prep, where you open the P&L and the summary before a check-in.

Current cost.

5 to 6 hr/wk x 50 weeks x $37 to $42/hr  =  $9,250 to $12,600 a year

What is recoverable. 30% at the low end, 50% at the high end. $2,775 to $6,300 a year, and 1.5 to 3.0 hours a week, all of them yours.

Why we are holding this below its ceiling, and it is a concession rather than caution. You told us you already use ChatGPT to draft client emails. Part of this gain is therefore already banked, and we are not going to sell it to you twice. This finding is High confidence and its ceiling is 80%. We are at 30% to 50% because the drafting-from-a-blank-page saving is largely yours already.

What is not banked, and this is the actual finding. Two things.

• Drafting from the numbers instead of from nothing. Today you open the P&L, read it, and write prose. The version that saves time starts from a short export of the month’s figures and the two or three movements worth mentioning, so the draft arrives with the content already in it and you are editing rather than composing.
• Doing it the same way thirty-eight times. Twenty-five minutes average across a range of ten minutes to an hour is a sign that each one is being approached fresh. A fixed shape for the 30 simpler clients is most of the win. The 8 harder ones should stay handwritten, because those are the two-entity, loan-carrying, question-asking clients and that is the paragraph they are paying you for.

And it doubles as your call prep. The hour a week you spend reading before a check-in is reading the same two artifacts. If the summary already leads with the two or three things that moved, the prep is done when the summary is.

What could, and where it should run. You told us you are on Google Workspace Business Starter, that Gemini shows up in Gmail now, and that you have not read the terms. We read them, and this matters more than the hours do.

Google’s position on Workspace data. Google states that interactions with Gemini in Workspace “stay within your organization”, that your content “is not shared with or used by any other customers”, and that submissions are “not used to train models and never reviewed by humans” under a Workspace with Gemini license. Your existing Workspace protections apply to it automatically. Gemini is included in all paid Workspace plans at no extra charge; the separate add-on was discontinued and base prices absorbed it. On Business Starter that means Gemini in the Gmail side panel, at a lower capability tier than the more expensive plans, for $0 more than you pay today.

And the honest version of the comparison, which is narrower than it sounds. For personal ChatGPT accounts on the Free, Plus and Pro plans, conversations are used to improve OpenAI’s models by default, and that default can be turned off at Settings, then Data Controls, then “Improve the model for everyone”. So the difference is not that one company uses your text and the other does not. The difference is that the Workspace protection is a term you have already accepted, applied automatically, with no per-person setting anyone can forget. The consumer protection depends on a toggle that starts in the wrong position and that each person has to find. In a two-person firm where one of you is on the account and one is not, that distinction is the whole thing.

Cost: $0. It is the mail system you already run the firm on.

The data boundary. Draft from a sanitized export: the month’s figures and movements, with the client’s name and account identifiers left out, which the draft does not need in order to be useful. Nothing goes to a client without you reading it, which is your rule and not our concession. And Google Workspace is not a new provider on your engagement letter’s list, so nothing here triggers the conversation you told us you would want to have first.

Verdict: habit. Nothing external is missing. The capability and the terms are already yours. Sources: Google’s Generative AI in Google Workspace Privacy Hub and its Workspace with Gemini business FAQ, and OpenAI’s Data Controls FAQ, all checked 2026-09-01. Confidence: high on the labor figures, which are your own hours.

3.4 New client details typed into three systems

You said it is fifteen minutes and you have never treated it as a priority. You were right.

What you told us. New client information goes into PandaDoc for the engagement letter, then into QuickBooks to open the company file, then into your Client Tracker sheet. “It’s maybe fifteen minutes per client so I’ve never treated it as a priority.” And you signed 11 new clients in the last twelve months.

Current cost.

11 new clients x 15 minutes = 2.75 hours a year x $37/hr = $101.75 a year

What is recoverable: nothing worth your attention. We are counting this at $0 and recommending you leave it exactly as it is. A hundred dollars a year is under three hours of your time annually. Any integration that removed it would take longer to set up than it will ever give back, and it would add a moving part to an onboarding that currently works. Your instinct on this was correct and it is the only finding in this report where the recommendation is to do nothing.

One thing in the same category that might not be small, and we do not have the number. When we asked Deshawn what he would add to your retyping list, he named two things you did not: client W-9s and vendor details typed into QuickBooks and then again into Gusto for the payroll clients, and the monthly Dext export matched by hand to QuickBooks on the four clients whose bank feeds lag. That second one is monthly rather than eleven times a year, and it is a matching job rather than a typing job, which usually means it is bigger than it looks.

Nobody has timed it, so it is not in any total in this report. Time it for one month and tell us during your question window and we will recompute at no charge. If it runs to even an hour a month across the four clients it outranks this finding by a factor of ten, and the fix is likely to be inside the same bank-rule work as Finding 3.2.

The data boundary. Your own onboarding, in your own systems. Nothing external.

Verdict: habit, and a cheap one. Leave it. Confidence: high. The arithmetic is small and simple and it does not depend on anything we could be wrong about.

3.5 What you actually pay for software

You said a couple hundred a month.

What your invoices say. We asked for them with plan names, and the plan names turned out to matter more than the amounts.

The split is the finding. $1,310 a month of that is client subscriptions you rebill inside each retainer. It moves through the firm and out again and it is not your cost. What you actually pay to run the firm is $416.80 a month, or $5,002 a year. You had it in your head as a couple hundred.

That is roughly double, and it is your discovery rather than ours: we only added up the invoices you already had. It is worth knowing for one specific reason. You told us your comfort for new tools is $100 to $250 a month total, and that ceiling is a lot easier to reason about once you know the base it sits on top of. The only new spend anywhere in this report is the optional $40 a month in Finding 3.1, which fits inside $250 with room.

There is no saving in this finding and we are not going to manufacture one. We counted it at $0. HubSpot and Mailchimp are both free, so cancelling them saves nothing and is a two-minute tidy-up rather than a finding. Here is everything we did find, with the arithmetic, none of it in any total:

• You are paying for ten Dext client slots that sit empty. Your tier covers 25 clients and you have 15 in it. At $255 that is $10.20 per slot across what you buy, or $17.00 per client across what you use. This is not a saving, it is an asset: the ten empty slots are why extending Dext to more clients in Finding 3.1 costs you nothing. Dext does not publish partner pricing, so we cannot tell you what a smaller tier would cost and we are not going to guess.
• PandaDoc, $35 a month for one Essentials seat on annual billing. The published Essentials annual price is around $19 a seat (checked 2026-09-01). We cannot see your account, so your invoice is what governs here and the gap is a question rather than a finding: it may be a legacy plan, an add-on, or a plan that was renamed underneath you. Ask them what the $35 is for. If it comes back as $19 that is $192 a year, which we have not counted because we do not know that it will.
• Calendly and Google Workspace are both on month-to-month billing. Calendly Standard is $12 monthly and $10 annually. Workspace Business Starter is $8.40 per user on flexible billing and $7 annually, and you have two users. Switching both to annual is $24 plus $33.60, so $57.60 a year. We have not counted it either, because it buys a saving with a twelve-month commitment and that is a trade rather than a free win.
• Financial Cents, paid and half used. You called this out yourself. It is not a saving at all, it is the delivery mechanism for Finding 3.1 and for the priority-list problem you named in your intake. The money is already spent. The question is whether the plan does what you think it does, which is in Finding 3.1.

The data boundary. Your own invoices. Nothing external.

Verdict: setting. The Dext capacity is bought and empty and the Financial Cents capability is bought and unused. Sources: your invoices dated 2026-09-01, plus PandaDoc, Calendly and Google Workspace published pricing checked the same day. Confidence: medium, because two of the four items above end in a question to a vendor rather than an answer from one.

3.6 The three or four more clients

The question you actually asked, answered in hours rather than in dollars.

What you told us. Given ten hours a week back: some of it for yourself, to stop working Saturday mornings during tax season, and the rest toward three or four more clients without hiring. A new client of the size you would want is $450 to $550 a month. Your book is 38 monthly clients at about $450 average, and 30 files get touched in a typical week.

We are counting this at $0. It is not a separate saving, it is what the first three findings buy if you spend them this way. Counting it again would be counting the same hours twice, and putting the biggest number in the audit on an assumption we cannot check would be the least honest thing in this document. So here is the arithmetic instead.

your current delivery load, from your own figures:
   categorization and reconciliation   27 hr/wk  (12 yours + 15 Deshawn’s)
   chasing                              8 hr/wk  (6 yours + 2 Deshawn’s)
   close summaries                      5 hr/wk
   call prep                            1 hr/wk
   payroll on Gusto                     2 hr/wk
                              total    43 hr/wk  across 38 clients
                                              =  1.13 hr/wk per client
   your share of that load: 24.7 of 43   =  57%

four more clients   4 x 1.13   =  4.5 hr/wk of firm time, of which about 2.6 is yours
what 3.1 + 3.2 + 3.3 return    =  7.0 to 13.4 hr/wk, of which 4.8 to 9.4 is yours
your Saturdays      12 x 4 hr  =  48 hr/yr, about 1.0 hr/wk averaged across the year

The answer, and it has two halves.

The half that works. At the low end you get back 4.8 hours a week of your own time, which is 240 hours a year. Four more clients need about 130 hours of your year and the Saturdays need 48. That is 178 against 240. On the annual average, the low end pays for both with about sixty hours to spare, and the high end is not close. Deshawn’s side works too, but only just: four clients need about 1.9 hours a week of his and the low end returns 2.2.

The half that does not. Those 48 Saturday hours are not spread across the year. They are four hours on each of twelve consecutive Saturdays between mid-January and April 15, and those are precisely the weeks when your volume peaks, when Deshawn goes to 30 hours, and when the recovered time is least available because everything else is at its heaviest. An annual average is the wrong unit for a seasonal problem. The hours are found across 50 weeks and the Saturdays need to be paid for inside 12 of them.

So the choice is real. If the four clients arrive in the autumn, they are consuming the same capacity you were going to spend on tax season, and February decides which one actually happened. Our suggestion is to sequence it: take the Saturdays back first, in the tax season after you do the work, and add clients from the spring once you have seen what the hours actually came to.

And the constraint that is not about total hours at all. Between a quarter and a third of everything found here is Deshawn’s time, not yours. New clients need your review, because your name and license are on the work and you told us that is not negotiable. So the binding limit on growth is your hours, not the firm’s, and Finding 3.3 matters more to this question than its dollar figure suggests: it is the one that returns time exclusively to you.

The revenue figure, stated and not claimed. Four clients at $450 a month is $21,600 a year. It is in this report once, here, in this sentence, and it is in no total anywhere. It requires you to go and sell four clients, which is a thing this audit does not do for you and which your close rate of about one in three on three proposals a month says would take a while.

The data boundary. Arithmetic on your own figures. Nothing external.

Verdict: not applicable. This is a consequence of the three findings above rather than a finding of its own. Confidence: medium, and it rests entirely on the hours figures you gave us.

4. What we deliberately did not pursue

Everything we looked at and left, so you can see it was decided rather than missed.

• Saturday work in tax season as its own finding. It is the consequence of Findings 3.1, 3.2 and 3.3, and counting it separately would count the same hours twice. It appears in 3.6 as the thing the hours buy back, which is how you framed it.
• Closes slipping past the tenth as its own finding. Same reason. You told us yourself it is caused by the chasing, so it is Finding 3.1’s consequence and it is already counted there.
• Moving off QuickBooks Online. You are a Certified ProAdvisor with 34 files on it. Migration would cost more than every finding in this report combined and nothing in your intake asks for it.
• Migrating the four Xero clients. A real annoyance, small money, and a client-relationship decision rather than an operations one. If they leave, let it happen at renewal.
• Any AI tool with access to client books. Your read-only answer was “maybe, tell me more”, and the more turned out to be that the two views worth having are your own QuickBooks Accountant view and your own Financial Cents view. We ruled out anything beyond that at the outline stage, not in the writing. Two ideas died there, and both would have looked good in this report.
• The Client Tracker sheet as a single point of failure. You flagged it yourself: “If that sheet disappeared I’d genuinely be stuck for a week.” That is a real risk and it is worth one sentence, which is this one. The honest fix is inside Financial Cents, which you already pay for, rather than a new tool, and it belongs after the thirty days in Section 6 rather than inside them.
• HubSpot and Mailchimp as growth levers. Finding 3.6 is about capacity to serve more clients, not about generating them. A marketing recommendation here would be us changing the subject.
• Anything above $250 a month. Your ceiling, and we treated it as hard rather than as a starting position.
• Answering routine “what does this mean” client emails, which is the third item on Deshawn’s list. It is real and we are not making it a finding, because it is the same quarter of the chase that Finding 3.1 already concedes it cannot automate: a client asking what a charge was for, or what a line on their P&L means, is a question that needs one of you. The part of it that is answerable in advance is the close summary, which is Finding 3.3, and a better summary is the only lever on this we would trust.
• The payroll work on Gusto. Nine clients, about two hours a week between you and Deshawn, and it costs the firm nothing because Gusto bills the clients directly. It is two hours of genuinely inherent work in the two days before each pay date. We looked and found nothing to remove.

Your three most-hated tasks, and where each one went. You named transaction categorization on the harder files (Finding 3.2), chasing clients (Finding 3.1), and writing close summaries (Finding 3.3). All three are findings with numbers. Nothing you named was quietly dropped.

5. What this is worth, and the bar we hold ourselves to

Total identified opportunity: 7.0 to 13.4 hours a week, and $12,070 to $25,540 a year.

New tool spend: $0 a year. Every counted finding runs on tools you already pay for. The one optional purchase in this report, the Financial Cents upgrade at $480 a year, is excluded from every figure above.

Net: $12,070 to $25,540 a year.

Even counting only what we’re certain of: $7,170/yr. That is Findings 3.2 and 3.3 alone, at the bottom of their ranges, which are the two we would stand behind with no benefit of the doubt because both rest on your own hours and neither depends on a client changing their behavior.

The bar we hold ourselves to. We don’t send an audit that finds less than $4,500 a year against a $1,500 fee. This one finds $12,070 at the bottom of its range, and $7,170 counting only the high-confidence findings. That is a standard for our own work, checked before you saw this, and it is the reason three of the six findings above are counted at zero rather than padded into the total.

What would move these numbers, in the order it would move them. Every figure here rests on hours you gave us in a follow-up conversation rather than on a system that measured them, and we would rather say that plainly than let the precision of the arithmetic imply a precision the inputs do not have.

1. The recurring-vendor share. We used two thirds, from your eyeball estimate on one file. It is the single number that moves this audit most.
2. Your hours on categorization. Twelve a week is the largest input in the report.
3. The Dext-to-QuickBooks matching nobody has timed. It could be larger than Finding 3.4.
4. Your Financial Cents plan’s actual reminder capability, which decides whether Finding 3.1 costs nothing or $40 a month.

Correct any of them during your question window and we recompute at no charge.

6. The first 30 days, in order

Free and fast first. Nothing in weeks 1 and 2 costs a dollar, and the one decision that costs anything waits until week 4, by which time you will have the evidence to make it.

Total cost for the thirty days: $0. Total return at the low end of the three findings the plan touches: $12,070 a year. The one paid decision sits at the end on purpose.

If you only do three things. Switch Dext Request paperwork on for the other eleven clients. Write the rules on the messiest file and import them into the other two. Check your Financial Cents plan page. That is under four hours, it costs nothing, and it is most of the low end.

Month two, once the wins are funding it. Give the Client Tracker sheet a second home inside Financial Cents, where the due dates and per-client notes already have somewhere to live. You told us you would be stuck for a week without it, and it is the one risk in this report that is not about hours.

7. Appendix: our assumptions

Rate assumptions. Two loaded rates, both sourced and dated, both disclosed against your own figures with the lower one governing.

Both rates in full. Your $37 to $42 is your own answer: a senior bookkeeper in Tulsa you would trust with the harder files is $28 to $32 an hour W-2, which burdens by 1.3365 to $37.42 to $42.77, rounded down at both ends. Deshawn’s $29 is BLS Occupational Employment and Wage Statistics, Oklahoma, Bookkeeping, Accounting and Auditing Clerks, mean $22.31 an hour (May 2024 survey); burdened by 1.3365 that is $29.82, rounded down.

Where the burden multiplier comes from. BLS Employer Costs for Employee Compensation, September 2024 reference period, private industry, establishments with 1 to 49 workers: total compensation $35.27 an hour against wages and salaries of $26.39. That ratio is 1.3365. We used the small-establishment cut rather than the all-industry one because your firm is two people, and because it is the lower of the two available.

Three disclosures about those rates, because they all push the numbers down.

1. A loaded rate normally means wage plus taxes plus benefits plus overhead. The federal series we used contains no overhead component at all. It covers wages, employer payroll taxes and employer-paid benefits and stops. So both rates above sit below a true loaded cost, and every dollar figure in this report is smaller than it would be on a fuller definition.
2. Deshawn’s own rate is higher than the one we used. He is W-2 at $24 an hour, which burdens to $32.08. The sourced replacement figure is $29.82. We used the lower one, which under-claims his hour by $3.08 against what you actually pay for it.
3. We cross-checked your rate a second way and it agreed. A 50/50 blend of the bookkeeping-clerk series and the accountant-and-auditor series for Oklahoma, both burdened, lands at $42.11, against the $42.77 top of your own band. Two independent methods within seventy cents is the best corroboration available here. We used the lower of them.

Your unpaid Saturdays are counted at the same rate as your paid hours, and that is a judgment call worth naming. You do not invoice a Saturday morning, so counting it at replacement cost prices unpaid owner overtime as though it were payroll. We counted it anyway, because the alternative is to value your own hours at less than everyone else’s, and that is the exact reasoning that produces Saturday mornings in the first place.

Working-week and volume assumptions. Fifty working weeks a year, which allows two weeks off and is the conservative choice: at fifty-two every annual figure here would be 4% larger.

Which volumes came from where. Everything marked “you told us” came from your intake form or your answers to our follow-up questions, and is your figure rather than ours. The transaction counts, the bank rule count and the Dext client counts came out of your own QuickBooks Accountant view and Financial Cents admin view under the read-only access you granted. The single most important number in this report, the two-thirds recurring-vendor share, is your eyeball estimate on one file, and it is labeled that way in Finding 3.2 rather than presented as measured.

What we did not measure and did not pretend to. Deshawn’s monthly Dext-to-QuickBooks matching on the four lagging clients. It is named in Finding 3.4 and it is in no total in this report.

Methodology. Findings were built from your intake form, a follow-up question set, and a research pass over the tools you named. Every finding carries a number and shows its calculation. Every capability and price claim was checked against the vendor’s own current documentation on 2026-09-01 and is dated in the finding that uses it. Where a vendor’s published plan table disagreed with what you told us, both are in the report and the disagreement is named rather than resolved silently. Where a vendor does not publish its pricing at all, which is the case for Dext’s partner tier, your invoice governs and we say we cannot compare it.

Confidence, and what it means here. High means the arithmetic rests on your own hours or your own system counts and does not depend on anyone changing behavior. Medium means something in it depends on a person or a vendor answer we do not have. Recoverable percentages are capped at 80% for high-confidence labor findings and 50% for medium-confidence ones. No finding in this report sits at its cap and three are counted at zero.

Delivery timeline. Day 0, intake received and parsed. Day 1, findings outline and the modeling decisions behind it, reviewed before any prose was written. Day 2, follow-up questions to you. Day 3, research against vendor documentation, every claim dated. Day 4, the numbers worksheet, where every figure in this report was computed and checked. Days 5 and 6, drafting and quality checks. Day 7, review and delivery.

Data handling. This report was produced for a fictional business created to demonstrate the work, so no client data of any kind was accessed. In a live audit, the read-only access you grant is used for counting only, it is never used to open a client file, and it is revoked on delivery. Working papers are deleted within thirty days and you get a confirmation when they are.

How to get help. Your questions are open by email for two weeks from the day this arrived. Reply to the delivery email and we answer. We check in on day ten in case something has come up that you have not asked about. And a live 30-minute readout call is included with your audit: book it with the link in the delivery email and we walk the findings together.

Bonus A: Getting named when a customer asks an AI assistant

More of your future clients will find you by asking an assistant than by typing into a search box, and the shape of that question is different. Nobody types “bookkeeper Tulsa” into an assistant. They type “I run a small construction company in Tulsa and I need someone to clean up two years of QuickBooks, who should I call”.

What we checked, and what we did not. We did not run assistant queries about your business, because your business does not exist: this is a fictional sample, and any assistant asked about Bramblewood Bookkeeping and Advisory would either say nothing or invent something. That is the honest expected result and reporting it is more useful than a fabricated screenshot. What follows is the input checklist, which is the part that is the same for every firm and the part you can act on. In a live audit this section carries a dated snapshot of what named assistants actually said, with the date and the number of questions stated on the page.

Who you are actually up against. Not the national franchises. When an assistant answers a question like the one above it tends to name two or three local firms with a real web presence, plus whichever directories it can read. In a market like Tulsa that is a short list, and being on it is mostly a matter of being legible rather than being loud.

Can the assistants even read your site? This is the part most firms fail and it is the cheapest to fix.

What to do about it. In order of return.

1. Check robots.txt. Five minutes. If it blocks assistant crawlers, unblocking is a one-line change and it is the difference between being unfindable and being findable.
2. Add LocalBusiness structured data naming the firm, the city, the services and the credential. You are a Certified ProAdvisor and that is exactly the kind of fact an assistant will repeat if it can read it and will omit if it cannot. Half an hour with a generator, or ask whoever built the site.
3. Write three answer-first pages for the three questions your best clients actually arrived with. QuickBooks cleanup, monthly bookkeeping for contractors, and getting your numbers ready for a loan application are the three your intake describes. Lead each with the answer in the first two sentences.
4. Make the name, address and phone identical everywhere, including the Google Business Profile.
5. Get reviews somewhere other than Google. Breadth of source matters more than volume on any one.

What this does and does not tell you. Assistant answers vary between runs and between people, so a single check is a snapshot rather than a measurement, and anyone who promises you placement in an assistant’s answer is selling something nobody can deliver. What the checklist above does is make you legible to the systems that are already reading. That is ordinary work with an ordinary payoff, and it is the honest version of this.

Bonus B: A starter AI policy

A starting point, not a legal document. Two people means you do not need much, but you do need the part that says what never goes into a chat box. Have your attorney read it before it becomes official, especially the client-data clause, since your engagement letters carry a confidentiality term this has to sit underneath.

AI Usage Policy, Bramblewood Bookkeeping and Advisory, LLC

1. Approved tools

2. What never goes into an AI tool that is not on the approved list

Client names. Employer identification numbers or Social Security numbers. Bank account or card numbers. Login details for anything. Payroll figures tied to a named person. Anything from a client that arrived marked confidential. If you are unsure, the answer is no and you ask Renata.

3. Client data

Client information may only be processed by providers named in the current engagement letter. Today that is QuickBooks, Dext and Gusto, and adding a provider means updating the list at renewal and telling the client, and nobody adds one without the owner’s approval.

4. Acceptable uses

Drafting anything a person will read and edit before it goes out. Summarizing documents the firm already holds. Explaining an accounting concept in plainer words. Checking your own reasoning.

5. Prohibited uses

Sending anything to a client that a person has not read first. Anything that reads as tax or legal advice. Letting a tool categorize transactions on a client file without review. Uploading a client document to any tool not on the approved list.

6. The review rule, which is the whole policy in one line

Final review before anything goes to a client. Always. The firm’s name and license are on the work, and no tool changes that.

7. If something goes wrong

Say so to Renata the same day. Not the next week and not after you have tried to fix it. Say what went in, to which tool, and when. Nobody is in trouble for reporting it; the only mistake that gets worse with time is the one nobody mentioned.

8. Review schedule

Reread this every six months and whenever a tool is added or a plan changes. Tool capabilities move faster than policies do, and a policy naming a feature that no longer works the way it is described is worse than no policy.

End of sample audit.